Nothing Special   »   [go: up one dir, main page]

WO2021226554A1 - Dual-band temperature detection systems and methods - Google Patents

Dual-band temperature detection systems and methods Download PDF

Info

Publication number
WO2021226554A1
WO2021226554A1 PCT/US2021/031435 US2021031435W WO2021226554A1 WO 2021226554 A1 WO2021226554 A1 WO 2021226554A1 US 2021031435 W US2021031435 W US 2021031435W WO 2021226554 A1 WO2021226554 A1 WO 2021226554A1
Authority
WO
WIPO (PCT)
Prior art keywords
image
visible
infrared
images
camera
Prior art date
Application number
PCT/US2021/031435
Other languages
French (fr)
Inventor
Austin A. Richards
Marcel Tremblay
Jeffrey D. Frank
Andres Prieto-Moreno
Original Assignee
Flir Systems Ab
Flir Systems Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Flir Systems Ab, Flir Systems Inc. filed Critical Flir Systems Ab
Publication of WO2021226554A1 publication Critical patent/WO2021226554A1/en
Priority to US17/981,337 priority Critical patent/US20230054197A1/en

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J5/0022Radiation pyrometry, e.g. infrared or optical thermometry for sensing the radiation of moving bodies
    • G01J5/0025Living bodies
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J5/02Constructional details
    • G01J5/08Optical arrangements
    • G01J5/0806Focusing or collimating elements, e.g. lenses or concave mirrors
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J5/10Radiation pyrometry, e.g. infrared or optical thermometry using electric radiation detectors
    • G01J5/20Radiation pyrometry, e.g. infrared or optical thermometry using electric radiation detectors using resistors, thermistors or semiconductors sensitive to radiation, e.g. photoconductive devices
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/21Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition
    • G06F18/20Analysing
    • G06F18/25Fusion techniques
    • G06F18/251Fusion techniques of input or preprocessed data
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/30Determination of transform parameters for the alignment of images, i.e. image registration
    • G06T7/33Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/70Determining position or orientation of objects or cameras
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/97Determining parameters from multiple pictures
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/10Image acquisition
    • G06V10/12Details of acquisition arrangements; Constructional details thereof
    • G06V10/14Optical characteristics of the device performing the acquisition or on the illumination arrangements
    • G06V10/143Sensing or illuminating at different wavelengths
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00Arrangements for image or video recognition or understanding
    • G06V10/70Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06VIMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00Scenes; Scene-specific elements
    • G06V20/50Context or environment of the image
    • G06V20/52Surveillance or monitoring of activities, e.g. for recognising suspicious objects
    • HELECTRICITY
    • H04ELECTRIC COMMUNICATION TECHNIQUE
    • H04NPICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N23/00Cameras or camera modules comprising electronic image sensors; Control thereof
    • H04N23/90Arrangement of cameras or camera modules, e.g. multiple cameras in TV studios or sports stadiums
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J2005/0077Imaging
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01JMEASUREMENT OF INTENSITY, VELOCITY, SPECTRAL CONTENT, POLARISATION, PHASE OR PULSE CHARACTERISTICS OF INFRARED, VISIBLE OR ULTRAVIOLET LIGHT; COLORIMETRY; RADIATION PYROMETRY
    • G01J5/00Radiation pyrometry, e.g. infrared or optical thermometry
    • G01J5/10Radiation pyrometry, e.g. infrared or optical thermometry using electric radiation detectors
    • G01J5/20Radiation pyrometry, e.g. infrared or optical thermometry using electric radiation detectors using resistors, thermistors or semiconductors sensitive to radiation, e.g. photoconductive devices
    • G01J2005/202Arrays
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10024Color image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/10Image acquisition modality
    • G06T2207/10048Infrared image
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]

Definitions

  • the present disclosure provides various embodiments of systems and methods for detecting temperature of an object in an image, such as measuring elevated body temperature (EBT) and/or elevated skin temperature (EST) for fever detection.
  • EBT elevated body temperature
  • EST elevated skin temperature
  • an infrared camera is used to measure skin temperature without physical contact with the subject, which is an important aspect of systems used for detecting potentially infectious diseases.
  • a person with a fever will have an elevated core body temperature and under certain conditions, this elevated temperature can manifest itself as elevated skin temperature, particularly on facial skin near the tear duct (canthus), an area of the face with a high degree of blood flow.
  • This canthus can be identified in an image and the measured surface temperature may correspond to the user’s body temperature (e.g., the canthus may be a few degrees cooler than the core body temperature).
  • a dual-band camera rig composed of a visible-light camera (e.g., producing an RGB image) and an infrared camera (e.g., a thermal IR camera) is used to acquire images that are spatially and temporally registered to a high degree of precision.
  • the RGB images are annotated manually and/or automatically (e.g., by human technicians, automatic object recognition software, etc.).
  • the annotations are then transferred to the infrared images, which themselves are much more difficult to annotate, since some edge details as well as color that gives context are not present in the thermal IR images.
  • FIG.1 is a flow chart of an example process for training a machine learning system for infrared image classification, in accordance with one or more embodiments of the present disclosure.
  • FIGs.2A, 2B, 2C and 2D illustrate differences between visible and infrared images, in accordance with one or more embodiments of the present disclosure.
  • FIG.3 illustrates a system for generating paired visible-light and infrared dataset images, in accordance with one or more embodiments of the present disclosure.
  • FIGs.4A, 4B, 4C, 4D and 4E are example visible and infrared image pairs captured using the system of FIG. 3, in accordance with one or more embodiments of the present disclosure.
  • FIG.5 illustrates an example infrared imaging and annotation system, in accordance with one or more embodiments of the present disclosure.
  • FIG.6A illustrates an infrared classification training system, in accordance with one or more embodiments.
  • FIG.6B illustrates an infrared classification system, in accordance with one or more embodiments.
  • FIGs.6C and 6D illustrate neural network training and validation processes, in accordance with one or more embodiments.
  • FIG.7 illustrates a relationship between blackbody temperature and in-band radiance, in accordance with one or more embodiments.
  • FIGs.8A and 8B illustrate a dual-band image capture system in accordance with one or more embodiments.
  • FIG.9 illustrates a thermal image with a test subject and two blackbodies, in accordance with one or more embodiments.
  • FIG.10 is a flow diagram illustrating a process for operating a thermal imaging system using one blackbody, in accordance with one or more embodiments.
  • FIG.11 is a figure illustrating a measurement of camera responsivity in ambient (23 C) operating conditions using two blackbodies.
  • FIG.12 is an example dual-camera system, in accordance with one or more embodiments.
  • FIG.13 is an example image captured with the dual-camera system of FIG.12, in accordance with one or more embodiments.
  • FIG.14 illustrates an example arrangement of a blackbody in a dual-band imaging system, in accordance with one or more embodiments.
  • FIGs.15A and 15B are example visible and infrared image pair, in accordance with one or more embodiments.
  • FIGs.16A, 16B, 16C and 16D are views of an example dual-band camera system, in accordance with one or more embodiments.
  • FIGs.17A and 17B are views of another example dual-band camera system, in accordance with one or more embodiments.
  • the present disclosure provides systems and methods for measuring temperature using a dual band image capture system.
  • a visible-light image of the subject for reference.
  • This reference image can be used to identify an object (e.g., a person to be measured), since it may be difficult to identify a person using a thermal image.
  • anatomical structures such as the inner corner of the eye (canthus) in both the visible and the thermal image when one is making a study of the hottest part of a person's face.
  • a dichroic beamsplitter is used to separate light by wavelength, making it possible to overlay visible and infrared image streams while minimizing parallax.
  • a dual-band training system will be first be described, followed by an example implementation of a trained dual image capture system.
  • Dual-Band Temperature Detection Training Systems and Methods The present disclosure provides systems and methods for annotating infrared images for use in machine learning applications.
  • a dual-band camera rig composed of a visible-light camera (e.g., producing an RGB image) and an infrared camera (e.g., a thermal IR camera) is used to acquire images that are spatially and temporally registered to a high degree of precision.
  • the RGB images are annotated manually and/or automatically (e.g., by human technicians, automatic object recognition software, etc.).
  • the annotations are then transferred to the infrared images, which themselves are much more difficult to annotate, since some edge details as well as color that gives context are not present in the thermal IR images.
  • CNN convolutional neural network
  • the images are labeled with annotations that are used in the training set.
  • the annotation is a process where a human operator manually labels objects in an image, for example pedestrians or stop signs in a traffic scene.
  • the CNN is trained with large datasets having thousands of images, for example.
  • the CNN is evaluated based on the correctness of its ability to identify the annotated objects on new images from a test dataset that were not part of the training set.
  • thermal IR images There is some difficulty, however, in having human operators correctly annotate thermal IR images. For example, a two-year old human being can quite easily identify many objects with a single glance, but even experts highly trained in the identification of objects in thermal images can have trouble deciding how to label objects in a scene.
  • the difficulties arise because many objects can and do look quite different in the thermal IR bands than they do in the visible band. For example, blue eyes do not look blue in a raw thermal image – the iris will be represented with shades of grey. Dark areas in a visible image can look quite bright in a thermal image and vice- versa, which adds to annotation confusion.
  • a thermal image can have an object with very high contrast between it and the scene, but very low contrast in the visible band.
  • An example process for training a machine learning system for infrared image classification in accordance with one or more embodiments will now be described with reference to FIG.1.
  • a process 100 includes, in step 110, capturing infrared images (e.g., thermal images) and visible-light images of the same scene at the same moment in time and from the same perspective.
  • step 120 the two images are registered so that a pixel on one image is registered to the corresponding pixel on the other image, and both pixels are viewing the same position in object space at the same moment in time.
  • a warping algorithm or other image alignment algorithm can be used.
  • the visible image and/or infrared image are classified.
  • automatic object classifier application can be run on the visible image and/or thermal image to identify object classifications, object locations, and other object information from the images.
  • a human operator reviews the preliminary classifications and decides how to annotate them.
  • the annotations can be limited to one image, such as the visible image which is more likely to be visually understandable to the user than the infrared image, and then applied to the other image, such as the infrared image.
  • the annotations may include an object location, an object classification and/or other object information.
  • the annotated infrared image is used in a dataset to train and/or validate a machine learning system for classifying an infrared image.
  • a convolutional neural network (CNN) is trained for image classification and validated to evaluate the CNN’s accuracy in classifying images and the objects in them, by using the annotated images to create a training dataset.
  • the annotation process as discussed above may be part of the CNN training process and may include manual classification where a human operator labels objects in the image by hand, for example pedestrians or stop signs.
  • human annotations may be assisted by running CNN detectors on the image that provide predictions (automatic or machine annotations) and, instead of starting from scratch, the annotator can then review the predictions and correct them if needed.
  • the CNN is trained with large datasets consisting of thousands of images, and the CNN is evaluated based on the correctness of its ability to identify the annotated objects on new images from a validation dataset that were not part of the training set.
  • a two-year old human being can quite easily identify many objects with a single glance, but even experts highly trained in the identification of objects in thermal images can have trouble deciding how to label objects in a scene.
  • blue eyes 202 in a visible image 200 do not look blue in a raw thermal image 210 – e.g., the iris may be represented with shades of grey 212 as illustrated.
  • dark areas 222 in a visible image 220 can look quite bright in a thermal image 230 and vice-versa, which adds to annotation confusion.
  • the TV 234 is barely discernable in the thermal image 230, just as the boy 232 is bright in the thermal image and barely apparent in the visible-light image.
  • a thermal image can have an object with very high contrast between it and the scene, but very low or no contrast in the visible band.
  • a visible image 240 and a midwave IR image 250 of hot footprints 252 illustrate this phenomenon.
  • the footprints 252 are not visibly identifiable in the visible image 240 but can be seen in the midwave IR image 250.
  • a solution is to train and teach the network for thermal image classification in the visible image and apply the annotation results to the thermal images in the training and validation datasets.
  • the solution includes both visible and thermal image classification, and the results are synthesized into the annotations as appropriate.
  • the datasets include groups of images for training and validating the neural models. The quality of the datasets affects the resulting models, with more accurate datasets providing better quality models.
  • the datasets may include multi-sensor datasets having images in two or more spectrums (e.g., visible, thermal, other spectrums).
  • image pairs are selected with accurate correspondence between the frames of the image pairs.
  • both spectrum frames may be taken at the same time from the substantially similar perspectives of the captured scene.
  • Another factor that affects training performance is the accuracy of the box or polygon (e.g., the ground truth) that identifies the object to be analyzed.
  • a beamsplitter-based camera is employed to with two cameras that are in sync, so frames are captured at the same time. Using this camera reduces the time needed to select and register frames for use in a dataset.
  • the system can select an appropriate data image from one of the spectrums (e.g., a high-quality image appropriate for a test image), and then the corresponding paired image will also be chosen for the dataset, thereby reducing time required to choose frames for the dataset.
  • an appropriate data image e.g., a high-quality image appropriate for a test image
  • annotations may be performed on the image that best represents (or shows) the desired object and those annotations can be matched to the image in the other spectrum. This method improves the accuracy of the classification because each spectrum be used to identify the borders of the object.
  • the multisensor camera may be combined in a system with a radiometric camera to provide additional functionally that would otherwise be difficult to implement with conventional systems.
  • Pixel-to-pixel alignment allows the CNN models to work in both spectrums to provide the best of each. For example, in an elevated body temperature system, well illuminated objects will go through CNN-visible networks to identify shapes and body parts, then the CNN-thermal networks may be used to identify in these parts the skin and measure the temperature of it. In another example, the system could use visible-light images to find the shape of an object (e.g., a car), and then use thermal images to locate and identify exhaust to determine whether a car is a electrical or gasoline powered vehicle. In some embodiments, having the pixel-to-pixel alignment allows a direct temperature measurement of each pixel location that is captured in the visible spectrum.
  • Visible-light CNN models can be used for detection of objects and/or objects-parts and the corresponding pixels from the thermal image can provide the temperature information.
  • This system has many advantages, including added privacy in systems in which the visible-light image is used only to facilitate accurate temperature measurements, and is displayed or available for use to further identify a person.
  • the visible image 270 on the right has been machine annotated, with blue bounding box 272 around a white car, but the car 262 is not annotated in the corresponding thermal image 260.
  • a solution, as disclosed herein, is to capture thermal images and visible-light images of the same scene at the same moment in time and from the same perspective.
  • a warping algorithm or other alignment algorithm can be used to register the two images so that a pixel on one channel is registered to the corresponding pixel on the other channel, and both these pixels are viewing the same position in object space at the same moment in time.
  • a human annotator can compare both images (and automatic image classification results, if available) and decide how to annotate them both.
  • the annotations for one image can then be applied to the other image.
  • This concept can be used to improve the auto-classification of thermal images using convolutional neural networks which have been trained with visible images. The visible-light image of the pair will be used for this matter, and results will be applicable to the corresponding thermal image.
  • Another advantage of the present embodiment is that classification can be performed with increased levels of privacy, because a third party can classify the images using visible image data without sharing the private infrared image data. The results can then be applied to the thermal IR images. For the scenes that only reveal context in thermal IR, a second phase may be applied that includes automated thermal image classification and/or human classification, but in general this technique will reduce the effort a lot.
  • FIG.3 Embodiments of a system for generating dataset images in accordance with the present disclosure are illustrated in FIG.3. The time synchronization can be accomplished by making one of the cameras trigger the other.
  • An infrared camera 320 (e.g., a high definition midwave infrared camera) is configured with synchronous outputs that can be routed into the sync input of a visible- light camera 310.
  • the two cameras are configured to start integration of a scene at the same time (e.g., within a few microseconds), with matched integration times which matches motion blur between the two channels.
  • Spatial registration can be accomplished with a system 300 that further includes a dichroic beamsplitter 302 which separates visible and infrared radiation from the scene.
  • the visible-light camera 310 and the infrared camera 320 are positioned to receive the separated visible and infrared radiation, respectively.
  • the infrared camera 320 is a midwave IR camera that captures a reflection 304 of the scene folded through 90 degrees of angle by the beamsplitter 302 which is viewed by the infrared camera 320 at a 45-degree angle.
  • the visible-light camera 310 looks through the beamsplitter 302 at the scene, which may include an object to detect and classify.
  • the illustrated system 300 eliminates all parallax.
  • the lenses for the two cameras may be selected to match the fields of view between the cameras, and/or adjustments to the captured images (e.g., cropping the larger image) may be performed after image capture.
  • the cameras are mounted on boresighting mounts 312 that can be adjusted in elevation and azimuth angles. Precision spacers underneath the visible-light camera mount sets the height of the visible- light optical axis above the optical breadboard to be the same as the midwave IR optical axis.
  • a glare stop 306, such as a piece of dark grey foam, is located to the left of the beamsplitter 302 in the illustrated embodiment to reduce reflections of visible-light off the beamsplitter side facing the visible-light camera 310.
  • the beamsplitter 302 may be implemented as a dichroic beamsplitter made of 0.3” BK7 glass. One side of the glass is coated with a layer of indium tin oxide (ITO) which makes it ⁇ 80% reflective in the midwave IR band (3-5 microns in wavelength units), and the other side has a built in visible-light anti-reflective coating.
  • ITO indium tin oxide
  • the beamsplitter 302 is also ⁇ 90% transparent to visible-light.
  • the beamsplitter 302 may be any beamsplitter that reflects a desired infrared band and is transparent to visible-light to allow for image capture at a quality that satisfies requirements of a particular system implementation. Referring to FIGs.
  • the two images can be spatially registered with high precision with various techniques, including a two-dimensional affine transform which can be defined by control points (such as control points 422 shown in visible image 420 and infrared image 430) that are common to the two images.
  • the affine transform warps the image to correct for differences in the lens distortions and results in precise image registration across the entire field of view.
  • the visible image 400 is warped to match the infrared image 410.
  • the control points 422 may be selected in places with high contrast in both images, as shown in visible image 420 and infrared image 430.
  • a registration process can include use of a target that has high- contrast fiducial points distributed over the entire field of view. This can be achieved with an ambient temperature white panel perforated with small holes and backlit with a black-painted heated panel.
  • the affine transform can be determined and used on subsequent images shot with the system, as long as the cameras and beamsplitter assembly all remain locked into the same relative positions. Referring to FIG.5, an example image capture system 500 that may be used to generate annotated infrared images as described herein will now be described, in accordance with one or more embodiments.
  • the image capture system is configured to capture and process both visible and infrared images, align the pair of captured images, and annotate the infrared image and/or visible image for use in training an image classifier for a variety of uses.
  • the image capture system 500 is configured to operate as described with respect to the embodiment of FIG. 3.
  • the image capture system 500 may be used for imaging a scene 570 in a field of view.
  • the image capture system 500 includes a processing component 510, a memory component 520, an infrared camera 501a, a visible-light camera 501b, an optional display component 540, a control component 550, a communication component 552, and other components depending on the system implementation.
  • the infrared camera 501a includes IR image optical components 532a (e.g., one or more lenses configured to receive radiation through an aperture 534a in infrared camera 501a and pass the radiation to IR image capture component 530a), IR image capture components 530a, and an IR image capture interface component 536a.
  • IR image capture components 530a include, in one embodiment, one or more sensors for capturing infrared image signals representative of an image, of scene 570.
  • the sensors of image capture components 530a provide for representing (e.g., converting) a captured infrared image signal of scene 570 as digital data (e.g., via an analog-to-digital converter included as part of the sensor or separate from the sensor as part of image capture system 500).
  • the image capture components 530a include infrared sensors (e.g., infrared detectors) implemented in an array or other fashion on a substrate.
  • infrared sensors may be implemented as a focal plane array (FPA).
  • FPA focal plane array
  • Infrared sensors may be configured to detect infrared radiation (e.g., infrared energy) from a target scene including, for example, midwave infrared wave bands (MWIR), longwave infrared wave bands (LWIR), and/or other thermal imaging bands as may be desired.
  • Infrared sensors may be implemented, for example, as microbolometers or other types of thermal imaging infrared sensors arranged in any desired array pattern to provide a plurality of pixels.
  • the infrared camera 501a may include a 3-5 micron high-definition MWIR camera (e.g., capturing infrared images at 1344x784 pixels).
  • the visible-light camera 501b includes visible image optical components 532b (e.g., one or more lenses configured to receive visible spectrum radiation through an aperture 534b in camera 501b and pass the received visible spectrum to visible image capture component 530b), visible image capture components 530b, and a visible image capture interface component 536b.
  • Visible image capture components 530b include, in one embodiment, one or more sensors for capturing visible-light image signals representative of an image of scene 570.
  • the sensors of visible image capture components 530b provide for representing (e.g., converting) a captured visible-light image signal of scene 570 as digital data (e.g., via an analog-to-digital converter included as part of the sensor or separate from the sensor as part of image capture system 500).
  • the visible image capture components 530b include light sensors implemented in an array or other fashion on a substrate.
  • sensors may be implemented as a charge-coupled-device (CCD) sensor, scientific complementary metal oxide semiconductor (sCMOS) sensor, or other visible-light sensor.
  • image capture system 500 may be implemented as a paired imaging system to simultaneously capture image frames of the scene 570 using IR camera 501a and visible- light camera 501b.
  • the cameras 501a and 501b may represent any type of camera system that is adapted to image the scene 570 and provide associated image data as described herein.
  • the image capture system 500 may be implemented at various types of fixed locations and environments, or in a portable device or vehicle.
  • the system includes a beamsplitter 502 which separates visible and infrared radiation from the scene 570.
  • the visible-light camera 501b and the infrared camera 501a are positioned to receive the separated visible and infrared radiation, respectively.
  • the infrared camera 501a (e.g., a midwave IR camera) is mounted to capture a reflection of the scene folded through a 90-degree angle by the beamsplitter 502 which is viewed by the infrared camera 501a at a 45-degree angle.
  • the visible-light camera 501b is mounted to capture a visible-light image of the scene 50 through the beamsplitter 502.
  • the optical components 532a and 532b for the two cameras may be selected to match the fields of view between the cameras, and/or adjustments to the captured images (e.g., cropping the larger image) may be performed after image capture.
  • the cameras are mounted on a board and can be adjusted in elevation and azimuth angles. Precision spacers underneath the mounts or other mounting components may be used to set the height of the optical axes to the same heights.
  • a glare stop 504 such as a piece of dark grey foam, is positioned adjacent to the beamsplitter 502 to reduce reflections of visible-light off the side of the beamsplitter 502 facing the visible-light camera 501b.
  • the beamsplitter 502 may be implemented as a dichroic beamsplitter made of 0.3” BK7 glass, with one side of the glass coated with a layer of indium tin oxide (ITO) which makes it ⁇ 80% reflective in the midwave IR band (3-5 microns in wavelength units).
  • ITO indium tin oxide
  • the beamsplitter 502 may be ⁇ 90% transparent to visible-light.
  • the beamsplitter 502 may be any beamsplitter that reflects a desired infrared band and is transparent to visible-light to allow for high quality image capture.
  • Processing component 510 may include, for example, a microprocessor, a single-core processor, a multi-core processor, a microcontroller, a logic device (e.g., a programmable logic device configured to perform processing operations), a digital signal processing (DSP) device, one or more memories for storing executable instructions (e.g., software, firmware, or other instructions), a graphics processing unit and/or any other appropriate combination of processing device and/or memory to execute instructions to perform any of the various operations described herein.
  • Processing component 510 is adapted to interface and communicate with components 536a, 536b, 520, 530, 540, and 550 to perform methods and processing steps as described herein.
  • Processing component 510 may also be adapted to perform synchronization 580 of the cameras 501a and 501b to capture images of the scene 570 at approximately the same time and with approximately the same integration period, image processing through image processing component 582, and/or image pair registration (image pair registration component 584) as described herein. It should be appreciated that processing operations and/or instructions may be integrated in software and/or hardware as part of processing component 510, or code (e.g., software or configuration data) which may be stored in memory component 520.
  • Embodiments of processing operations and/or instructions disclosed herein may be stored by a machine-readable medium in a non-transitory manner (e.g., a memory, a hard drive, a compact disk, a digital video disk, or a flash memory) to be executed by one or more computers (e.g., logic or processor-based system) to perform various methods disclosed herein.
  • Memory component 520 includes, in one embodiment, one or more memory devices (e.g., one or more memories) to store data and information.
  • the one or more memory devices may include various types of memory including volatile and non-volatile memory devices, such as RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically-Erasable Read- Only Memory), flash memory, or other types of memory.
  • processing component 510 is adapted to execute software stored in memory component 520 and/or a machine- readable medium to perform various methods, processes, and operations in a manner as described herein.
  • Processing component 510 may be adapted to receive image signals from image capture components 530a and 530b, process image signals (e.g., to provide processed image data), store image signals or image data in memory component 520, and/or retrieve stored image signals from memory component 520.
  • processing component 510 may be remotely positioned, and processing component 510 may be adapted to remotely receive image signals from image capture components 530 via wired or wireless communication with image capture interface component 536, as described herein.
  • Display component 540 may include an image display device (e.g., a liquid crystal display (LCD)) or various other types of generally known video displays or monitors.
  • Control component 550 may include, in various embodiments, a user input and/or interface device, such as a keyboard, a control panel unit, a graphical user interface, or other user input/output.
  • Control component 550 may be adapted to be integrated as part of display component 540 to operate as both a user input device and a display device, such as, for example, a touch screen device adapted to receive input signals from a user touching different parts of the display screen.
  • Processing component 510 may be adapted to communicate with image capture interface components 536a and 536b (e.g., by receiving data and information from image capture components 530a and 530b).
  • Image capture interface components 536a and 536b may be configured to receive image signals (e.g., image frames) from image capture components 530a and 530b, respectively, and communicate image signals to processing component 510 directly or through one or more wired or wireless communication components (e.g., represented by connection 537) in the manner of communication component 552.
  • communication component 552 may be implemented as a network interface component adapted for communication with a network and may include one or more wired or wireless communication components.
  • a network may be implemented as a single network or a combination of multiple networks, and may include a wired or wireless network, including a wireless local area network, a wide area network, the Internet, a cloud network service, and/or other appropriate types of communication networks.
  • the image capture system 500 may be configured to operate with one or more computing devices, servers and/or one or more databases, and may be combined with other components.
  • image capture system 500 may send image pairs over a network (e.g., the Internet or the cloud) to a server system, for remote image pair registrations and processing, annotations, and other processes as disclosed herein. Registered image pairs may be provided to an annotation system 556 for further processing.
  • the annotation system 556 may be integrated into a local computing system with one or more other components of image capture system 500, accessed through a wireless or wired communications link, accessed through a network, such as the Internet or a cloud service, a standalone system (e.g., receiving registered image pairs via an external memory device), a mobile system, or other system configured to perform the systems and methods described herein.
  • the annotation system 556 includes infrared classification components 557 for automatically (e.g., using a trained CNN) and/or manually (e.g., user interface) analyzing infrared images and visible classification components 558 for automatically and/or manually analyzing visible images.
  • the image classification may include, for example, detecting one or more objects in an image, defining a bounding box around detected objects, and/or classifying detected objects.
  • the annotation components 559 are configured to provide an interface that synthesizes the infrared and visible classification information for an image pair allowing a user to view the images and proposed annotations and confirm and/or edit the annotations.
  • the annotated image pairs may then be stored in a database 560 for use in training and/or validating a neural network for infrared image classification.
  • Various aspects of the present disclosure may be implemented for training neural networks and/or other machine learning processes to analyze and/or classify captured infrared images for a variety of applications, including surveillance, traffic monitoring, detection and tracking of people, fever monitoring, etc.
  • Embodiments of neural networking training systems and methods that may be used in the present disclosure will now be described with reference to FIGs. 6A-D. Referring to FIG.6A, embodiments of an infrared classification system will be described.
  • the infrared classification system 600 may be implemented as a standalone system and/or on one or more servers such as an application server that performs data processing and/or other software execution operations for training, storing, and neural networks used by the infrared classification system 600.
  • the components of the infrared classification system 600 may be distributed across a communications network, such as cloud/network 622.
  • the communications network 622 may include one or more local networks such as a wireless local area network (WLAN), wide area networks such as the Internet or cloud network, and other wired or wireless communications paths suitable for facilitating communications between components as described herein.
  • WLAN wireless local area network
  • the infrared classification system 600 includes communications components 614 operable to facilitate communications with one or more remote systems, such as a remote infrared device 624 configured to capture one or more infrared images of a scene and detect and/or classify objects therein, and an infrared/visible image capture system 620 configured to capture registered infrared/visible image pairs for use in training an infrared image classification system.
  • the infrared classification system 600 may operate as a networked infrared image classification system, such as a cloud-based system, or may be configured to operate in a dedicated system, such as a surveillance system that processes thermal images and other data captured in real time from one or more surveillance devices (e.g., a thermal imaging camera as described herein).
  • the infrared classification system 600 may be configured to analyze the captured data and return information relating to an infrared image classification, such as location of detection objects, classification of detected objects, confidence measure for the classification, etc.
  • the infrared classification system 600 may also include a database 602 for storing captured infrared/visible image pairs, training datasets, trained neural networks, and other information.
  • the infrared classification system 600 includes one or more processors 604 that perform data processing and/or other software execution operations.
  • the processor 604 may include logic devices, microcontrollers, processors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other devices that may be used by the infrared image classification system 600 to execute appropriate instructions, such as software instructions stored in memory 606 including image pair classification and annotation components 608, infrared classification training system components 610, trained infrared classification neural networks 612 (e.g., a convolutional neural network trained by a training dataset stored in the database 602), and/or other applications.
  • the memory 606 may be implemented in one or more memory devices (e.g., memory components) that store executable instructions, data and information, including image data, video data, audio data, network information.
  • the memory devices may include various types of memory for information storage including volatile and non-volatile memory devices, such as RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically-Erasable Read-Only Memory), flash memory, a disk drive, and other types of memory described herein.
  • the remote infrared device 624 may be implemented as a computing device such as a thermal imaging camera, a handheld temperature sensing device, a desktop computer or network server, a mobile computing device such as a mobile phone, tablet, laptop computer or other computing device having communications circuitry (e.g., wireless communications circuitry or wired communications circuitry) for connecting with other devices.
  • the remote infrared device 624 may include one or more unmanned vehicles (e.g., drones) such as an unmanned aerial vehicle, an unmanned ground vehicle, or other unmanned vehicle.
  • the communications components 614 may include circuitry for communicating with other devices using various communications protocols.
  • communications components 614 may be configured to communicate over a wired communication link (e.g., through a network router, switch, hub, or other network devices) for wired communication purposes.
  • a wired link may be implemented with a power-line cable, a coaxial cable, a fiber-optic cable, or other appropriate cables or wires that support corresponding wired network technologies.
  • Communications components 614 may be further configured to interface with a wired network and/or device via a wired communication component such as an Ethernet interface, a power-line modem, a Digital Subscriber Line (DSL) modem, a Public Switched Telephone Network (PSTN) modem, a cable modem, and/or other appropriate components for wired communication.
  • a wired communication component such as an Ethernet interface, a power-line modem, a Digital Subscriber Line (DSL) modem, a Public Switched Telephone Network (PSTN) modem, a cable modem, and/or other appropriate components for wired communication.
  • DSL Digital Subscriber Line
  • PSTN Public Switched Telephone Network
  • cable modem a cable modem
  • communications components 614 may be further configured to interface with a wired network and/or device via a wired communication component such as an Ethernet interface, a power-line modem, a Digital Subscriber Line (DSL) modem, a Public Switched Telephone Network (PSTN) modem, a
  • the infrared classification system 650 includes a processor and memory 660, operable to store one or more trained neural networks and implement the neural network run-time interface (such as trained infrared classification neural network 670) thereon.
  • a training dataset stored in the database 602 may be created from annotated registered image pairs and used to train one or more neural networks and other machine learning algorithms for use in an infrared classification system.
  • FIG. 6C an embodiment of a neural network training process will now be described.
  • the neural network 680 is a convolutional neural network (CNN) that receives the training dataset 682 and outputs a classification for the data.
  • CNN convolutional neural network
  • the present disclosure describes a plurality of neural networks that may be trained for one or more infrared image determinations, including but not limited to, detecting and tracking objects and/or people from thermal images, detection of a temperature measurement location in a thermal image, classification of a detected object, and/or determination of whether an individual has a fever.
  • the training dataset includes annotated infrared image data created from registered visual and infrared pairs as described herein.
  • the data may also include annotated infrared data created through other means, including synthetic data generated to simulate real-world images.
  • the training starts with a forward pass through the neural network 680 including feature extraction, a plurality of convolution layers and pooling layers, a plurality of fully connected layers, and an output layer that includes the desired classification.
  • a backward pass through the neural network 680 may be used to update the CNN parameters in view of errors produced in the forward pass (e.g., misclassified data).
  • other processes may be used in accordance with the present disclosure.
  • An embodiment for validating the trained neural network is illustrated in FIG.6D.
  • a set of fully annotated validation test data 692 is fed into the trained neural network 690.
  • the validation test data 692 may include annotated infrared data generated from registered infrared/visible image pairs as described herein, that was not used as part of the training dataset 682.
  • Detected errors may be analyzed and fed back to the training system to update the training model, which in turn updates the training dataset 682 to create a more accurate classification model.
  • detected errors may be corrected by adding more examples of the data (e.g., more types of environments), increasing the resolution of the data and/or increasing the accuracy of the thermal modeling, to help distinguish between data types.
  • the training dataset By adjusting the training dataset to improve accuracy on-the-fly, the operator can avoid costly delays in implementing accurate classification systems.
  • the system is configured to save data generated in real-time in the field for analysis and training of one or more neural networks.
  • data from a deployed system may be fed back into a CNN training process to refine the CNN to improve classification for a particular environment (e.g., detect temperature of people in an airport), desired classification goal (e.g., train the CNN to detect and track one or more objects) and/or for more accurate performance.
  • desired classification goal e.g., train the CNN to detect and track one or more objects
  • dual-Band Temperature Detection Systems and Methods The various training processes disclosed herein may be used to train infrared imaging systems for a variety of detection and classification tasks.
  • systems for measuring elevated body temperature (EBT) make use of thermal infrared cameras to measure skin temperature without physical contact--an important aspect of any system used in the presence of people with potentially infectious diseases.
  • a person with a fever will have an elevated core body temperature and under certain conditions, this elevated temperature can manifest itself as elevated skin temperature, particularly on facial skin near the tear duct (canthus), an area of the face with a high degree of blood flow.
  • This canthus surface temperature may only be a few degrees C cooler than the core body temperature.
  • the thermal infrared camera is calibrated to accurately measure the difference between a healthy person and a febrile person may only be a matter of a few degrees C difference in core body temperature.
  • the standard calibration of an infrared camera relates the digital counts for each pixel in the image to the radiance of a surface viewed by those pixels (in the spectral band of the thermal camera).
  • the temperature of the surface can be derived from the radiance measured by the camera by various methods, including a lookup table. In the case of skin temperature measurements, it is possible to calibrate the system directly from digital counts to temperature for two reasons. The first reason is that the range of facial skin temperatures is limited to values from 30 °C to ⁇ 38 °C. Over this range of temperature, the in-band radiance for a 7-14 micron longwave camera is nearly linear in temperature, as shown in FIG 7. Referring to FIGs.8A and 8B, an example embodiment of a trained dual-band imaging system will be described, in accordance with one or more embodiments.
  • a dual-band camera 800 includes a first camera 810, a second camera 820, and a beamsplitter 830.
  • the first camera 810 includes image capture and processing components configured to sense and capture a visible image of a field of view.
  • the second camera 820 includes image capture and processing components configured to sense and capture an infrared image of the field of view.
  • the first camera 810 and the second camera 820 are arranged in a camera body 802 and configured to capture the images through a beamsplitter 830 at substantially the same time.
  • the beamsplitter 830 reflects an image of the scene to the second camera 820, while the first camera 810 captures the visible images through the beamsplitter 830.
  • the arrangement disclosed herein allows image pairs to be captured with minimal parallax, which facilitates accurate image processing and classification as described herein.
  • a blackbody 840 may be present in the field of view.
  • the blackbody may be positioned as part of the dual-band camera 800 and/or positioned closed to the camera body 820.
  • the reflection of the blackbody 840 radiation off the beamsplitter 830 allows the black body to positioned physically closer to the thermal camera due to the folded geometry (e.g., black body 840 is (a) + (b) away from the thermal camera 820).
  • placing blackbody close to the thermal camera allows the black body to appear larger to the thermal camera, and also allows for precision placement of the blackbody at a location to be captured by the thermal camera 820, but outside of the usable image.
  • the blackbody may include a thermistor, a micro-cavity blackbody (e.g., a long, tapered cone) or other blackbodies.
  • a thermistor a calibrated device with a resistance value as a function of temperature
  • FIG.9 The skin on a person’s face has an in-band emissivity that is very close to the typical emissivity of an area blackbody source, e.g., 0.96.
  • FIG.9 illustrates a thermal image 900 of a test subject 902 with a dual blackbody system in the field of view.
  • a first blackbody 910 has a temperature of 35 C
  • a second blackbody 912 has a temperature of 41 C.
  • Temperatures can be interpolated and extrapolated from the digital count values on the two blackbodies, or on a single blackbody if the responsivity of the camera is known. In the case of the single blackbody configuration, it may be necessary to know the responsivity of the camera. Referring to FIG.10, a method 1000 to measure temperature using a single reference blackbody will now be described in accordance with one or more embodiments.
  • the responsivity of the camera is measured using two blackbodies (e.g., at 35C and 40C) with the camera fully warmed up and operating in an environment representative of an expected temperature of operation.
  • the camera software may be configured to provide frame averaging to reduce temporal noise, and the responsivity may be calculated as the difference in digital counts on two regions of interest (ROIs) positioned on the two blackbodies (e.g., 571.8 counts), divided by the temperature difference, which in this example is 5° C.
  • the resulting responsivity is 114 counts/°C.
  • a region of interest is identified on the reference blackbody in the field of view of the camera while viewing a scene of interest which contains test subjects, and the mean digital counts on the reference blackbody are determined. This value is the reference offset.
  • the system measures the digital counts on the target of interest, which may include a region on the face of a test subject (e.g., the hottest part of the face, a specific target location on the face such as a the canthus).
  • the software could automatically detect the person’s face, then locate the hottest spot on the face (e.g., the canthus) and place a spotmeter with a 3x3 pixel size on the centroid of the hot spot to generate the EBT counts measurement.
  • a visible images is captured and used to identify the object and/or location on the object, and the identified location can then be measured at the corresponding location on the thermal image
  • the reference offset is subtracted from the EBT counts measurement.
  • the system then divides by the responsivity to get a relative temperature difference between the test subject’s canthus and the reference blackbody. This temperature difference is then added to the blackbody temperature.
  • the blackbody temperature could be read out continuously by the software to have an updated measurement. This measurement value could be inserted into the metadata of the thermal image.
  • a visible-light image of the test subject is recorded which could be used to further analyze the scene (e.g., help identify the subject). Additional metadata could also be recorded, including the air temperature in the proximity of the measurement site, the relative humidity, the range to the person from the camera (perhaps measured by a stereo camera rig, LIDAR, RADAR or another method).
  • a thermal image 1100 is captured by the infrared imaging camera of a field of view.
  • the thermal image 1100 includes an image of a blackbody 1120, which has an associated region of interest (e.g., defined by green box labeled 35C Reference) in the thermal image 1100, from which temperature measurements are taken.
  • a target 1110 is identified in the thermal image 1110 has an associated region of interest (e.g., defined by red box labeled 40C Target) from which temperature measurements for the target are taken.
  • the dual-band camera 1200 is configured to reduce system parallax using a dichroic beamsplitter 1220.
  • the dual-band camera 1200 may operate with a trained neural network that was trained using a dual-band camera system and beamsplitter as discussed herein. In the operation, the duel-band camera system 1200 captures video data from an infrared camera 1230 and video data from a visible camera 1210 simultaneously with the optical axes overlaid at a high level of precision.
  • the dual-band camera 1200 uses the beamsplitter 1220 at a 45-degree angle relative to the optical axis of each camera (visible camera 1210 and infrared camera 1230).
  • the beamsplitter 1230 is coated with indium tin oxide on one side to reflect LWIR radiation with a ⁇ 90% efficiency to the infrared camera 1230, while simultaneously transmitting 90% of visible-light to the visible camera 1210.
  • the illustrated system could be implemented in a variety of thermal imaging applications, including EBT applications because there is negligible parallax error between the visible and thermal IR images at all distances.
  • a conventional system with two side-by-side image sensors e.g., boresighted visible and thermal IR cameras side- by-side
  • a blackbody 1240 is located in a corner of the IR field of view (or other location that does not obstruct or minimally obstructs the captured image).
  • the dataset could be presented as a visible video of a target person where the system user could mouse over any point on the visible image and see the temperature on each point. This embodiment allows a system operator to intuitively explore different ways to view temperature in an EBT or other temperature measurement implementation.
  • the visible image may be discarded after object detection, analysis, and temperature measurement is performed to preserve the privacy of the subject (e.g., as may be required by law).
  • the dual-band camera 1200 is controlled by a processing component 1250 (e.g., such as one or more of the processing components of FIGs.5-6B), configured in accordance with a particular use or implementation including synchronization logic 1260 for synchronizing the image capture of the two cameras (e.g., synchronizing shutters and image acquisition for temporal registration of captured images), image processing logic 1262 for processing the captured images into images for use by the system, image pair registration logic 1264 for aligning the image pair (e.g., pixel-by-pixel correspondence), object detection and tracking logic 1266 for identifying one or more target objects in the image pair, and temperature measurement logic 1268 for determining a temperature of an object or portion of an object.
  • synchronization logic 1260 for synchronizing the image capture of the two cameras (e.g., synchronizing shutters and image acquisition for temporal registration of captured
  • one or more of the logical components may be implemented using image analysis, trained neural networks, statistical analysis, and/or other logic.
  • the dual-band camera 1200 communicates with a remote processing system 1272 through communications components 1270.
  • the remote processing system 1272 may provide one or more services for the dual-band camera 1200 including training and downloading neural networks, online processing of image pairs, storage of historical data, and other services.
  • An an example of a method of displaying EBT data taken with the beamsplitter system is illustrated in FIG.13.
  • a display 1300 includes the visible image 1310 of an object (e.g., a person), which is overlaid with a thermal image inset 1320 on the visible image 1310 to show temperature data at a selected location (e.g., a location selected by the user by placing a pointer 1330 over a desired region).
  • a selected location e.g., a location selected by the user by placing a pointer 1330 over a desired region.
  • the hottest part of the test subject’s face is the yellow portion on the canthus.
  • the temperature is 33.1 °C as shown by the color bar on the right side.
  • the concept of placing a reference blackbody in the scene near the test subject or at the same distance as the test subject has the advantage that atmospheric transmission losses are “basked into” the calibration because both the test subject and the reference blackbody have their infrared signals changed by the same amount: reduced by the transmission of the air path, and increased by the air path signals themselves.
  • placing the reference blackbody in the scene at distance has some distinct disadvantages. For one, the camera cannot be panned or tilted over the scene unless the reference blackbody is kept in the field of view and its location is known so that the software can maintain the positioning of an ROI on it.
  • a system 1400 addresses these issues by placing a reference blackbody 1410 in the field of view of a thermal imaging device 1430 at a close distance to the camera lens.
  • the blackbody 1410 includes a small circular emitter made of aluminum coated with a high emissivity coating. A heating coil would be used to heat up the emitter and a thermistor or thermocouple could be used to monitor the absolute temperature of the emitter.
  • the blackbody 1410 will be highly out of focus if the blackbody is centimeters from the lens of the camera, while the lens is focused at some convenient working distance (e.g., 3 meters or some similar distance where everything in the scene from 1 meter to infinity would be in sufficient focus for measurement). If the blackbody 1410 is out of focus, then it will not be perceived as being at its actual physical temperature because it is a small target and the radiance from it will be smeared out over more pixels than if it was in focus.
  • a solution to this problem is to place a collimating lens 1420 in front of the blackbody emitter 1410 to focus its rays so that the blackbody appears to be in focus.
  • a shutter is periodically placed in front of the IR camera lens to expose the sensor to a known temperature scene.
  • the shutter could be instrumented with temperature sensors which would then be used to calculate the reference offset.
  • the shutter could also serve as the flat field correction shutter which would improve the spatial uniformity of the images.
  • Putting the flat field correction shutter in front of the lens also corrects for the non- uniformity introduced by the lens assembly.
  • the disadvantage of this approach is that temperature measurements of test subjects would not have the reference blackbody in the field of view at the same time, reducing traceability.
  • systems and methods are provided for correcting for distance- induced temperature errors.
  • the system is configured to correct the apparent temperature of a head based on its distance from the thermal imaging sensors so that someone who is sick does not get counted as healthy just because they are farther away from the camera.
  • One is atmospheric transmission.
  • the other is the so-called “spot size effect”.
  • spot size effect manifests itself as a decreasing apparent temperature of a target with decreasing angular size. It is observed that, with respect to the modulation transfer function of the optical system, as a hot target gets smaller, there is a “mixing” of cold pixels at the edges that begins to reduce the apparent temperature of the target.
  • the spot size effect is pronounced when the target is falls below a threshold size (e.g., is less than 15 by 15 pixels in size).
  • the canthus hot spot on an adult is about 1cm by 1cm. It is observed that for an adult person’s canthus to be at least 15x15 pixels in an image formed by an example camera having a 640 x 512 resolution with a 32-degree horizontal field of view, the target person should be less than 0.8 meters away. For screening of people walking by in a crowd, it may be necessary to measure the distance from the camera to the heads of the targets, perhaps based on the size of the head (assuming adult heads).
  • the distance measurement is used in an air path transmission model that accounts for target distance, air temp (e.g., using FPA temperature or external temperature sensor) and relative humidity (e.g., using a relative humidity sensor).
  • the air path transmission model is based on Moderate Resolution Atmospheric Transmission (MODTRAN), a radiative transport model developed by the Air Force.
  • MODTRAN Moderate Resolution Atmospheric Transmission
  • Image pair 1500 illustrates an example of a visible-light image 1510 showing fog and a thermal image 1520, showing an image of a man 1522 in the fog.
  • Image pair 1550 illustrates an example of a visible-light image 1560 showing a person behind an obstruction 1570, and a corresponding thermal infrared image 1570.
  • These examples illustrate benefits of a dual image camera system as described herein for both training and detection.
  • the image pairs illustrate different contexts where labeling of visible and infrared images provide different context information for the scene. Object detection, tracking and classification can be improved using both visible-light images and infrared images, such as tracking an object when the visible-light image is obstructed.
  • FIGs.16A, 16B, 16C and 16D are views of an example dual-band camera system, in accordance with one or more embodiments, including a perspective cutaway view (FIG.16A), a side cutaway view (FIG. 16B), a bottom view (FIG.16C), and a perspective view (FIG.16D).
  • An embodiment of a dual-band camera system 1600 includes a housing 1602, an infrared camera 1610, a beamsplitter 1620, a visible-light camera 1630 and optical components 1632 (e.g., a lens assembly) for focusing the visible-light image towards the image capture components of the visible- light camera 1630.
  • optical components 1632 e.g., a lens assembly
  • FIGs.17A and 17B provide views of another example dual-band camera system, in accordance with one or more embodiments, including a plan view (FIG. 17A) and a front side view (FIG. 17B).
  • the dual-band camera system 1700 includes a housing 1720 with a cover 1712 enclosing the system components, which include an infrared camera 1710 (e.g., a LWIR camera), a beam splitter 1720, a visible-light camera 1730 and processing components 1740 (e.g., a GPU board).
  • the dual-band camera system 1700 is configured to receive infrared waves 1706 and visible- light waves 1708 of a scene.
  • the beamsplitter 1720 allows the visible-light waves 1708 to pass through to the visible-light camera 1730.
  • the visible-light camera 1730 includes optical components and/or separate optical components 1732 are provided to focus the visible-light 1708 for the image capture components of the visible-light camera 1730.
  • the beamsplitter 1720 is also configured to reflect the infrared waves 1706 through window 1712 towards the infrared camera 1710.
  • Processing components 1740 are provided to warp the infrared images to overlay on the visible-light images and generate a two-channel, dual band video stream.
  • various embodiments provided by the present disclosure can be implemented using hardware, software, or combinations of hardware and software.
  • the various hardware components and/or software components set forth herein can be combined into composite components comprising software, hardware, and/or both without departing from the spirit of the present disclosure.
  • the various hardware components and/or software components set forth herein can be separated into sub-components comprising software, hardware, or both without departing from the spirit of the present disclosure.
  • Software in accordance with the present disclosure such as non-transitory instructions, program code, and/or data, can be stored on one or more non-transitory machine-readable mediums. It is also contemplated that software identified herein can be implemented using one or more general purpose or specific purpose computers and/or computer systems, networked and/or otherwise.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Human Computer Interaction (AREA)
  • Computing Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Databases & Information Systems (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Software Systems (AREA)
  • Signal Processing (AREA)
  • Image Processing (AREA)
  • Studio Devices (AREA)
  • Image Analysis (AREA)

Abstract

Systems and methods include an infrared camera configured to capture an infrared image of a scene, a visible-light camera configured to capture a visible-light image of the scene, and a logic device configured to simultaneously capture a pair of images of the scene comprising the infrared image of the scene and the visible image of the scene, align the pair of images so that a pixel location in one of the pair of images has a corresponding pixel location m the other image, classify the visible image, annotate the infrared image based, at least in part, on the classification of the visible image, and add the annotated infrared image to a neural network training dataset for use in training a neural network for infrared image classification. A. beamsplitter is arranged to refl ect a first image of the scene towards the infrared camera and pass through a second image of the scene to the visible-light camera, and a first b!ackbodv is attached thereto and positioned in a field of view of the infrared camera.

Description

DUAL-BAND TEMPERATURE DETECTION SYSTEMS AND METHODS Austin A. Richards, Marcel Tremblay, Jeffrey D. Frank, and Andres Prieto-Moreno CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to and the benefit of U.S. Provisional Patent Application No. 63/022,281 filed May 8, 2020 and entitled “DUAL-BAND TEMPERATURE DETECTION SYSTEMS AND METHODS,” which is hereby incorporated by reference in its entirety. TECHNICAL FIELD One or more embodiments of the present disclosure relate generally to thermal imaging systems, and more particularly, for example, to systems and methods for detecting elevated body temperature and/or elevated skin temperature using a dual-band images. BACKGROUND In the field of image processing, there is an ongoing need for efficient and reliable ways to analyze and process images captured by imaging devices. Conventional systems may include machine learning systems trained on labeled image datasets. However, training and validating these systems are subject to error when the images are difficult to accurately label. For example, thermal images present difficulties because human operators, even experts highly trained in the identification of objects in thermal images, can have trouble deciding how to label objects in a thermal scene. The difficulties arise because many objects can and do look quite different in the thermal IR bands than they do in the visible band. In view of the foregoing, there is a continued need in the art for improved image processing systems and methods. SUMMARY The present disclosure provides various embodiments of systems and methods for detecting temperature of an object in an image, such as measuring elevated body temperature (EBT) and/or elevated skin temperature (EST) for fever detection. In some embodiments, an infrared camera is used to measure skin temperature without physical contact with the subject, which is an important aspect of systems used for detecting potentially infectious diseases. A person with a fever will have an elevated core body temperature and under certain conditions, this elevated temperature can manifest itself as elevated skin temperature, particularly on facial skin near the tear duct (canthus), an area of the face with a high degree of blood flow. This canthus can be identified in an image and the measured surface temperature may correspond to the user’s body temperature (e.g., the canthus may be a few degrees cooler than the core body temperature). Various systems and methods are provided for annotating infrared images for use in machine learning applications. In some embodiments, a dual-band camera rig composed of a visible-light camera (e.g., producing an RGB image) and an infrared camera (e.g., a thermal IR camera) is used to acquire images that are spatially and temporally registered to a high degree of precision. The RGB images are annotated manually and/or automatically (e.g., by human technicians, automatic object recognition software, etc.). The annotations are then transferred to the infrared images, which themselves are much more difficult to annotate, since some edge details as well as color that gives context are not present in the thermal IR images. The scope of the disclosure is defined by the claims, which are incorporated into this section by reference. A more complete understanding of embodiments of the invention will be afforded to those skilled in the art, as well as a realization of additional advantages thereof, by a consideration of the following detailed description of one or more embodiments. Reference will be made to the appended sheets of drawings that will first be described briefly. BRIEF DESCRIPTION OF THE DRAWINGS FIG.1 is a flow chart of an example process for training a machine learning system for infrared image classification, in accordance with one or more embodiments of the present disclosure. FIGs.2A, 2B, 2C and 2D illustrate differences between visible and infrared images, in accordance with one or more embodiments of the present disclosure. FIG.3 illustrates a system for generating paired visible-light and infrared dataset images, in accordance with one or more embodiments of the present disclosure. FIGs.4A, 4B, 4C, 4D and 4E are example visible and infrared image pairs captured using the system of FIG. 3, in accordance with one or more embodiments of the present disclosure. FIG.5 illustrates an example infrared imaging and annotation system, in accordance with one or more embodiments of the present disclosure. FIG.6A illustrates an infrared classification training system, in accordance with one or more embodiments. FIG.6B illustrates an infrared classification system, in accordance with one or more embodiments. FIGs.6C and 6D illustrate neural network training and validation processes, in accordance with one or more embodiments. FIG.7 illustrates a relationship between blackbody temperature and in-band radiance, in accordance with one or more embodiments. FIGs.8A and 8B illustrate a dual-band image capture system in accordance with one or more embodiments. FIG.9 illustrates a thermal image with a test subject and two blackbodies, in accordance with one or more embodiments. FIG.10 is a flow diagram illustrating a process for operating a thermal imaging system using one blackbody, in accordance with one or more embodiments. FIG.11 is a figure illustrating a measurement of camera responsivity in ambient (23 C) operating conditions using two blackbodies. FIG.12 is an example dual-camera system, in accordance with one or more embodiments. FIG.13 is an example image captured with the dual-camera system of FIG.12, in accordance with one or more embodiments. FIG.14 illustrates an example arrangement of a blackbody in a dual-band imaging system, in accordance with one or more embodiments. FIGs.15A and 15B are example visible and infrared image pair, in accordance with one or more embodiments. FIGs.16A, 16B, 16C and 16D are views of an example dual-band camera system, in accordance with one or more embodiments. FIGs.17A and 17B are views of another example dual-band camera system, in accordance with one or more embodiments. Embodiments of the disclosure and their advantages are best understood by referring to the detailed description that follows. It should be appreciated that like reference numerals are used to identify like elements illustrated in one or more of the figures. DETAILED DESCRIPTION The present disclosure provides systems and methods for measuring temperature using a dual band image capture system. When one is measuring elevated body temperature using thermal imaging, it may be useful to have a visible-light image of the subject for reference. This reference image can be used to identify an object (e.g., a person to be measured), since it may be difficult to identify a person using a thermal image. It is also useful for the correlation of anatomical structures such as the inner corner of the eye (canthus) in both the visible and the thermal image when one is making a study of the hottest part of a person's face. In various embodiments, a dichroic beamsplitter is used to separate light by wavelength, making it possible to overlay visible and infrared image streams while minimizing parallax. In the disclosure provided herein, a dual-band training system will be first be described, followed by an example implementation of a trained dual image capture system. Dual-Band Temperature Detection Training Systems and Methods The present disclosure provides systems and methods for annotating infrared images for use in machine learning applications. In some embodiments, a dual-band camera rig composed of a visible-light camera (e.g., producing an RGB image) and an infrared camera (e.g., a thermal IR camera) is used to acquire images that are spatially and temporally registered to a high degree of precision. The RGB images are annotated manually and/or automatically (e.g., by human technicians, automatic object recognition software, etc.). The annotations are then transferred to the infrared images, which themselves are much more difficult to annotate, since some edge details as well as color that gives context are not present in the thermal IR images. In order to properly train a convolutional neural network (CNN) for image classification and evaluate the CNN’s ability to classify images and the objects in them, the images are labeled with annotations that are used in the training set. In many applications, the annotation is a process where a human operator manually labels objects in an image, for example pedestrians or stop signs in a traffic scene. The CNN is trained with large datasets having thousands of images, for example. The CNN is evaluated based on the correctness of its ability to identify the annotated objects on new images from a test dataset that were not part of the training set. There is some difficulty, however, in having human operators correctly annotate thermal IR images. For example, a two-year old human being can quite easily identify many objects with a single glance, but even experts highly trained in the identification of objects in thermal images can have trouble deciding how to label objects in a scene. The difficulties arise because many objects can and do look quite different in the thermal IR bands than they do in the visible band. For example, blue eyes do not look blue in a raw thermal image – the iris will be represented with shades of grey. Dark areas in a visible image can look quite bright in a thermal image and vice- versa, which adds to annotation confusion. In some cases, a thermal image can have an object with very high contrast between it and the scene, but very low contrast in the visible band. An example process for training a machine learning system for infrared image classification in accordance with one or more embodiments will now be described with reference to FIG.1. In various embodiments, a process 100 includes, in step 110, capturing infrared images (e.g., thermal images) and visible-light images of the same scene at the same moment in time and from the same perspective. In step 120, the two images are registered so that a pixel on one image is registered to the corresponding pixel on the other image, and both pixels are viewing the same position in object space at the same moment in time. In various embodiments, a warping algorithm or other image alignment algorithm can be used. In step 130, the visible image and/or infrared image are classified. In some embodiments, automatic object classifier application can be run on the visible image and/or thermal image to identify object classifications, object locations, and other object information from the images. In step 140, a human operator reviews the preliminary classifications and decides how to annotate them. In some embodiments, the annotations can be limited to one image, such as the visible image which is more likely to be visually understandable to the user than the infrared image, and then applied to the other image, such as the infrared image. The annotations may include an object location, an object classification and/or other object information. In step 150, the annotated infrared image is used in a dataset to train and/or validate a machine learning system for classifying an infrared image. In some embodiments, a convolutional neural network (CNN) is trained for image classification and validated to evaluate the CNN’s accuracy in classifying images and the objects in them, by using the annotated images to create a training dataset. The annotation process, as discussed above may be part of the CNN training process and may include manual classification where a human operator labels objects in the image by hand, for example pedestrians or stop signs. In some embodiments, human annotations may be assisted by running CNN detectors on the image that provide predictions (automatic or machine annotations) and, instead of starting from scratch, the annotator can then review the predictions and correct them if needed. The CNN is trained with large datasets consisting of thousands of images, and the CNN is evaluated based on the correctness of its ability to identify the annotated objects on new images from a validation dataset that were not part of the training set. As previously discussed, there is some difficulty in having human operators correctly annotate thermal IR images. A two-year old human being can quite easily identify many objects with a single glance, but even experts highly trained in the identification of objects in thermal images can have trouble deciding how to label objects in a scene. The difficulties arise because many objects can and do look quite different in the thermal IR bands than they do in the visible band. For example, as illustrated in FIG.2A, blue eyes 202 in a visible image 200 do not look blue in a raw thermal image 210 – e.g., the iris may be represented with shades of grey 212 as illustrated. Referring to FIG.2B, dark areas 222 in a visible image 220 can look quite bright in a thermal image 230 and vice-versa, which adds to annotation confusion. Here the TV 234 is barely discernable in the thermal image 230, just as the boy 232 is bright in the thermal image and barely apparent in the visible-light image. In some cases, a thermal image can have an object with very high contrast between it and the scene, but very low or no contrast in the visible band. Referring to FIG.2C, for example, a visible image 240 and a midwave IR image 250 of hot footprints 252 illustrate this phenomenon. In this example, the footprints 252 are not visibly identifiable in the visible image 240 but can be seen in the midwave IR image 250. In various embodiments disclosed herein, a solution is to train and teach the network for thermal image classification in the visible image and apply the annotation results to the thermal images in the training and validation datasets. In some embodiments, the solution includes both visible and thermal image classification, and the results are synthesized into the annotations as appropriate. The datasets include groups of images for training and validating the neural models. The quality of the datasets affects the resulting models, with more accurate datasets providing better quality models. The datasets may include multi-sensor datasets having images in two or more spectrums (e.g., visible, thermal, other spectrums). When building datasets for multi-sensor training systems, image pairs are selected with accurate correspondence between the frames of the image pairs. As described herein, both spectrum frames may be taken at the same time from the substantially similar perspectives of the captured scene. Another factor that affects training performance is the accuracy of the box or polygon (e.g., the ground truth) that identifies the object to be analyzed. In some embodiments, a beamsplitter-based camera is employed to with two cameras that are in sync, so frames are captured at the same time. Using this camera reduces the time needed to select and register frames for use in a dataset. The system can select an appropriate data image from one of the spectrums (e.g., a high-quality image appropriate for a test image), and then the corresponding paired image will also be chosen for the dataset, thereby reducing time required to choose frames for the dataset. After the accurate alignment of the two images (e.g., alignment of the pixels of each image), annotations may be performed on the image that best represents (or shows) the desired object and those annotations can be matched to the image in the other spectrum. This method improves the accuracy of the classification because each spectrum be used to identify the borders of the object. In some embodiments, the multisensor camera may be combined in a system with a radiometric camera to provide additional functionally that would otherwise be difficult to implement with conventional systems. Pixel-to-pixel alignment allows the CNN models to work in both spectrums to provide the best of each. For example, in an elevated body temperature system, well illuminated objects will go through CNN-visible networks to identify shapes and body parts, then the CNN-thermal networks may be used to identify in these parts the skin and measure the temperature of it. In another example, the system could use visible-light images to find the shape of an object (e.g., a car), and then use thermal images to locate and identify exhaust to determine whether a car is a electrical or gasoline powered vehicle. In some embodiments, having the pixel-to-pixel alignment allows a direct temperature measurement of each pixel location that is captured in the visible spectrum. Visible-light CNN models can be used for detection of objects and/or objects-parts and the corresponding pixels from the thermal image can provide the temperature information. This system has many advantages, including added privacy in systems in which the visible-light image is used only to facilitate accurate temperature measurements, and is displayed or available for use to further identify a person. Referring to FIG.2D, the visible image 270 on the right has been machine annotated, with blue bounding box 272 around a white car, but the car 262 is not annotated in the corresponding thermal image 260. A solution, as disclosed herein, is to capture thermal images and visible-light images of the same scene at the same moment in time and from the same perspective. A warping algorithm or other alignment algorithm can be used to register the two images so that a pixel on one channel is registered to the corresponding pixel on the other channel, and both these pixels are viewing the same position in object space at the same moment in time. A human annotator can compare both images (and automatic image classification results, if available) and decide how to annotate them both. The annotations for one image can then be applied to the other image. This concept can be used to improve the auto-classification of thermal images using convolutional neural networks which have been trained with visible images. The visible-light image of the pair will be used for this matter, and results will be applicable to the corresponding thermal image. Another advantage of the present embodiment is that classification can be performed with increased levels of privacy, because a third party can classify the images using visible image data without sharing the private infrared image data. The results can then be applied to the thermal IR images. For the scenes that only reveal context in thermal IR, a second phase may be applied that includes automated thermal image classification and/or human classification, but in general this technique will reduce the effort a lot. Embodiments of a system for generating dataset images in accordance with the present disclosure are illustrated in FIG.3. The time synchronization can be accomplished by making one of the cameras trigger the other. An infrared camera 320 (e.g., a high definition midwave infrared camera) is configured with synchronous outputs that can be routed into the sync input of a visible- light camera 310. The two cameras are configured to start integration of a scene at the same time (e.g., within a few microseconds), with matched integration times which matches motion blur between the two channels. Spatial registration can be accomplished with a system 300 that further includes a dichroic beamsplitter 302 which separates visible and infrared radiation from the scene. The visible-light camera 310 and the infrared camera 320 are positioned to receive the separated visible and infrared radiation, respectively. In the illustrated embodiment, the infrared camera 320 is a midwave IR camera that captures a reflection 304 of the scene folded through 90 degrees of angle by the beamsplitter 302 which is viewed by the infrared camera 320 at a 45-degree angle. The visible-light camera 310 looks through the beamsplitter 302 at the scene, which may include an object to detect and classify. In various embodiments, it is desirable for the optical axes of the visible-light camera 310 and the infrared camera 320 to be precisely lined up so that there is negligible perspective change between the two, which allows for the creation of accurately registered images at all distances. If the two cameras are mounted next to each other, there will be a parallax error between them. The illustrated system 300 eliminates all parallax. The lenses for the two cameras may be selected to match the fields of view between the cameras, and/or adjustments to the captured images (e.g., cropping the larger image) may be performed after image capture. In some embodiments, the cameras are mounted on boresighting mounts 312 that can be adjusted in elevation and azimuth angles. Precision spacers underneath the visible-light camera mount sets the height of the visible- light optical axis above the optical breadboard to be the same as the midwave IR optical axis. A glare stop 306, such as a piece of dark grey foam, is located to the left of the beamsplitter 302 in the illustrated embodiment to reduce reflections of visible-light off the beamsplitter side facing the visible-light camera 310. As illustrated, the beamsplitter 302 may be implemented as a dichroic beamsplitter made of 0.3” BK7 glass. One side of the glass is coated with a layer of indium tin oxide (ITO) which makes it ~80% reflective in the midwave IR band (3-5 microns in wavelength units), and the other side has a built in visible-light anti-reflective coating. The beamsplitter 302 is also ~90% transparent to visible-light. In various embodiments, the beamsplitter 302 may be any beamsplitter that reflects a desired infrared band and is transparent to visible-light to allow for image capture at a quality that satisfies requirements of a particular system implementation. Referring to FIGs. 4A-E, a pair of images taken with an example system are shown. The two images (such as visible image 400 and infrared image 410) can be spatially registered with high precision with various techniques, including a two-dimensional affine transform which can be defined by control points (such as control points 422 shown in visible image 420 and infrared image 430) that are common to the two images. The affine transform warps the image to correct for differences in the lens distortions and results in precise image registration across the entire field of view. In one approach, the visible image 400 is warped to match the infrared image 410. The control points 422 may be selected in places with high contrast in both images, as shown in visible image 420 and infrared image 430. The resulting overlay image 440 of the two images is shown with a color “map” with the “fixed” midwave IR image shown in green and the moving image shown in purple. In some embodiments, a registration process can include use of a target that has high- contrast fiducial points distributed over the entire field of view. This can be achieved with an ambient temperature white panel perforated with small holes and backlit with a black-painted heated panel. The affine transform can be determined and used on subsequent images shot with the system, as long as the cameras and beamsplitter assembly all remain locked into the same relative positions. Referring to FIG.5, an example image capture system 500 that may be used to generate annotated infrared images as described herein will now be described, in accordance with one or more embodiments. In the illustrated embodiment, the image capture system is configured to capture and process both visible and infrared images, align the pair of captured images, and annotate the infrared image and/or visible image for use in training an image classifier for a variety of uses. In some embodiments, the image capture system 500 is configured to operate as described with respect to the embodiment of FIG. 3. The image capture system 500 may be used for imaging a scene 570 in a field of view. The image capture system 500 includes a processing component 510, a memory component 520, an infrared camera 501a, a visible-light camera 501b, an optional display component 540, a control component 550, a communication component 552, and other components depending on the system implementation. The infrared camera 501a includes IR image optical components 532a (e.g., one or more lenses configured to receive radiation through an aperture 534a in infrared camera 501a and pass the radiation to IR image capture component 530a), IR image capture components 530a, and an IR image capture interface component 536a. IR image capture components 530a include, in one embodiment, one or more sensors for capturing infrared image signals representative of an image, of scene 570. The sensors of image capture components 530a provide for representing (e.g., converting) a captured infrared image signal of scene 570 as digital data (e.g., via an analog-to-digital converter included as part of the sensor or separate from the sensor as part of image capture system 500). In some embodiments, the image capture components 530a include infrared sensors (e.g., infrared detectors) implemented in an array or other fashion on a substrate. For example, infrared sensors may be implemented as a focal plane array (FPA). Infrared sensors may be configured to detect infrared radiation (e.g., infrared energy) from a target scene including, for example, midwave infrared wave bands (MWIR), longwave infrared wave bands (LWIR), and/or other thermal imaging bands as may be desired. Infrared sensors may be implemented, for example, as microbolometers or other types of thermal imaging infrared sensors arranged in any desired array pattern to provide a plurality of pixels. In some embodiments, the infrared camera 501a may include a 3-5 micron high-definition MWIR camera (e.g., capturing infrared images at 1344x784 pixels). The visible-light camera 501b includes visible image optical components 532b (e.g., one or more lenses configured to receive visible spectrum radiation through an aperture 534b in camera 501b and pass the received visible spectrum to visible image capture component 530b), visible image capture components 530b, and a visible image capture interface component 536b. Visible image capture components 530b include, in one embodiment, one or more sensors for capturing visible-light image signals representative of an image of scene 570. The sensors of visible image capture components 530b provide for representing (e.g., converting) a captured visible-light image signal of scene 570 as digital data (e.g., via an analog-to-digital converter included as part of the sensor or separate from the sensor as part of image capture system 500). In some embodiments, the visible image capture components 530b include light sensors implemented in an array or other fashion on a substrate. For example, sensors may be implemented as a charge-coupled-device (CCD) sensor, scientific complementary metal oxide semiconductor (sCMOS) sensor, or other visible-light sensor. In various embodiments, image capture system 500 may be implemented as a paired imaging system to simultaneously capture image frames of the scene 570 using IR camera 501a and visible- light camera 501b. In various embodiments, the cameras 501a and 501b may represent any type of camera system that is adapted to image the scene 570 and provide associated image data as described herein. The image capture system 500 may be implemented at various types of fixed locations and environments, or in a portable device or vehicle. The system includes a beamsplitter 502 which separates visible and infrared radiation from the scene 570. The visible-light camera 501b and the infrared camera 501a are positioned to receive the separated visible and infrared radiation, respectively. The infrared camera 501a (e.g., a midwave IR camera) is mounted to capture a reflection of the scene folded through a 90-degree angle by the beamsplitter 502 which is viewed by the infrared camera 501a at a 45-degree angle. The visible-light camera 501b is mounted to capture a visible-light image of the scene 50 through the beamsplitter 502. In various embodiments, it is desirable for the optical axes of the visible-light camera 501b and the infrared camera 501a to be precisely lined up so that there is negligible perspective change between the two captured images, which allows for the creation of accurately registered images at various distances. The optical components 532a and 532b for the two cameras may be selected to match the fields of view between the cameras, and/or adjustments to the captured images (e.g., cropping the larger image) may be performed after image capture. In some embodiments, the cameras are mounted on a board and can be adjusted in elevation and azimuth angles. Precision spacers underneath the mounts or other mounting components may be used to set the height of the optical axes to the same heights. A glare stop 504, such as a piece of dark grey foam, is positioned adjacent to the beamsplitter 502 to reduce reflections of visible-light off the side of the beamsplitter 502 facing the visible-light camera 501b. The beamsplitter 502 may be implemented as a dichroic beamsplitter made of 0.3” BK7 glass, with one side of the glass coated with a layer of indium tin oxide (ITO) which makes it ~80% reflective in the midwave IR band (3-5 microns in wavelength units). In this embodiment, the beamsplitter 502 may be ~90% transparent to visible-light. In various embodiments, the beamsplitter 502 may be any beamsplitter that reflects a desired infrared band and is transparent to visible-light to allow for high quality image capture. Processing component 510 may include, for example, a microprocessor, a single-core processor, a multi-core processor, a microcontroller, a logic device (e.g., a programmable logic device configured to perform processing operations), a digital signal processing (DSP) device, one or more memories for storing executable instructions (e.g., software, firmware, or other instructions), a graphics processing unit and/or any other appropriate combination of processing device and/or memory to execute instructions to perform any of the various operations described herein. Processing component 510 is adapted to interface and communicate with components 536a, 536b, 520, 530, 540, and 550 to perform methods and processing steps as described herein. Processing component 510 may also be adapted to perform synchronization 580 of the cameras 501a and 501b to capture images of the scene 570 at approximately the same time and with approximately the same integration period, image processing through image processing component 582, and/or image pair registration (image pair registration component 584) as described herein. It should be appreciated that processing operations and/or instructions may be integrated in software and/or hardware as part of processing component 510, or code (e.g., software or configuration data) which may be stored in memory component 520. Embodiments of processing operations and/or instructions disclosed herein may be stored by a machine-readable medium in a non-transitory manner (e.g., a memory, a hard drive, a compact disk, a digital video disk, or a flash memory) to be executed by one or more computers (e.g., logic or processor-based system) to perform various methods disclosed herein. Memory component 520 includes, in one embodiment, one or more memory devices (e.g., one or more memories) to store data and information. The one or more memory devices may include various types of memory including volatile and non-volatile memory devices, such as RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically-Erasable Read- Only Memory), flash memory, or other types of memory. In one embodiment, processing component 510 is adapted to execute software stored in memory component 520 and/or a machine- readable medium to perform various methods, processes, and operations in a manner as described herein. Processing component 510 may be adapted to receive image signals from image capture components 530a and 530b, process image signals (e.g., to provide processed image data), store image signals or image data in memory component 520, and/or retrieve stored image signals from memory component 520. In various aspects, processing component 510 may be remotely positioned, and processing component 510 may be adapted to remotely receive image signals from image capture components 530 via wired or wireless communication with image capture interface component 536, as described herein. Display component 540 may include an image display device (e.g., a liquid crystal display (LCD)) or various other types of generally known video displays or monitors. Control component 550 may include, in various embodiments, a user input and/or interface device, such as a keyboard, a control panel unit, a graphical user interface, or other user input/output. Control component 550 may be adapted to be integrated as part of display component 540 to operate as both a user input device and a display device, such as, for example, a touch screen device adapted to receive input signals from a user touching different parts of the display screen. Processing component 510 may be adapted to communicate with image capture interface components 536a and 536b (e.g., by receiving data and information from image capture components 530a and 530b). Image capture interface components 536a and 536b may be configured to receive image signals (e.g., image frames) from image capture components 530a and 530b, respectively, and communicate image signals to processing component 510 directly or through one or more wired or wireless communication components (e.g., represented by connection 537) in the manner of communication component 552. In one or more embodiments, communication component 552 may be implemented as a network interface component adapted for communication with a network and may include one or more wired or wireless communication components. In various embodiments, a network may be implemented as a single network or a combination of multiple networks, and may include a wired or wireless network, including a wireless local area network, a wide area network, the Internet, a cloud network service, and/or other appropriate types of communication networks. The image capture system 500 may be configured to operate with one or more computing devices, servers and/or one or more databases, and may be combined with other components. In some embodiments, image capture system 500 may send image pairs over a network (e.g., the Internet or the cloud) to a server system, for remote image pair registrations and processing, annotations, and other processes as disclosed herein. Registered image pairs may be provided to an annotation system 556 for further processing. In various embodiments, the annotation system 556 may be integrated into a local computing system with one or more other components of image capture system 500, accessed through a wireless or wired communications link, accessed through a network, such as the Internet or a cloud service, a standalone system (e.g., receiving registered image pairs via an external memory device), a mobile system, or other system configured to perform the systems and methods described herein. In various embodiments, the annotation system 556 includes infrared classification components 557 for automatically (e.g., using a trained CNN) and/or manually (e.g., user interface) analyzing infrared images and visible classification components 558 for automatically and/or manually analyzing visible images. The image classification may include, for example, detecting one or more objects in an image, defining a bounding box around detected objects, and/or classifying detected objects. The annotation components 559 are configured to provide an interface that synthesizes the infrared and visible classification information for an image pair allowing a user to view the images and proposed annotations and confirm and/or edit the annotations. The annotated image pairs may then be stored in a database 560 for use in training and/or validating a neural network for infrared image classification. Various aspects of the present disclosure may be implemented for training neural networks and/or other machine learning processes to analyze and/or classify captured infrared images for a variety of applications, including surveillance, traffic monitoring, detection and tracking of people, fever monitoring, etc. Embodiments of neural networking training systems and methods that may be used in the present disclosure will now be described with reference to FIGs. 6A-D. Referring to FIG.6A, embodiments of an infrared classification system will be described. The infrared classification system 600 may be implemented as a standalone system and/or on one or more servers such as an application server that performs data processing and/or other software execution operations for training, storing, and neural networks used by the infrared classification system 600. In some embodiments, the components of the infrared classification system 600 may be distributed across a communications network, such as cloud/network 622. The communications network 622 may include one or more local networks such as a wireless local area network (WLAN), wide area networks such as the Internet or cloud network, and other wired or wireless communications paths suitable for facilitating communications between components as described herein. The infrared classification system 600 includes communications components 614 operable to facilitate communications with one or more remote systems, such as a remote infrared device 624 configured to capture one or more infrared images of a scene and detect and/or classify objects therein, and an infrared/visible image capture system 620 configured to capture registered infrared/visible image pairs for use in training an infrared image classification system. In various embodiments, the infrared classification system 600 may operate as a networked infrared image classification system, such as a cloud-based system, or may be configured to operate in a dedicated system, such as a surveillance system that processes thermal images and other data captured in real time from one or more surveillance devices (e.g., a thermal imaging camera as described herein). The infrared classification system 600 may be configured to analyze the captured data and return information relating to an infrared image classification, such as location of detection objects, classification of detected objects, confidence measure for the classification, etc. The infrared classification system 600 may also include a database 602 for storing captured infrared/visible image pairs, training datasets, trained neural networks, and other information. As illustrated, the infrared classification system 600 includes one or more processors 604 that perform data processing and/or other software execution operations. The processor 604 may include logic devices, microcontrollers, processors, application specific integrated circuits (ASICs), field programmable gate arrays (FPGAs) or other devices that may be used by the infrared image classification system 600 to execute appropriate instructions, such as software instructions stored in memory 606 including image pair classification and annotation components 608, infrared classification training system components 610, trained infrared classification neural networks 612 (e.g., a convolutional neural network trained by a training dataset stored in the database 602), and/or other applications. The memory 606 may be implemented in one or more memory devices (e.g., memory components) that store executable instructions, data and information, including image data, video data, audio data, network information. The memory devices may include various types of memory for information storage including volatile and non-volatile memory devices, such as RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically-Erasable Read-Only Memory), flash memory, a disk drive, and other types of memory described herein. The remote infrared device 624 may be implemented as a computing device such as a thermal imaging camera, a handheld temperature sensing device, a desktop computer or network server, a mobile computing device such as a mobile phone, tablet, laptop computer or other computing device having communications circuitry (e.g., wireless communications circuitry or wired communications circuitry) for connecting with other devices. In some embodiments, the remote infrared device 624 may include one or more unmanned vehicles (e.g., drones) such as an unmanned aerial vehicle, an unmanned ground vehicle, or other unmanned vehicle. The communications components 614 may include circuitry for communicating with other devices using various communications protocols. In various embodiments, communications components 614 may be configured to communicate over a wired communication link (e.g., through a network router, switch, hub, or other network devices) for wired communication purposes. For example, a wired link may be implemented with a power-line cable, a coaxial cable, a fiber-optic cable, or other appropriate cables or wires that support corresponding wired network technologies. Communications components 614 may be further configured to interface with a wired network and/or device via a wired communication component such as an Ethernet interface, a power-line modem, a Digital Subscriber Line (DSL) modem, a Public Switched Telephone Network (PSTN) modem, a cable modem, and/or other appropriate components for wired communication. Proprietary wired communication protocols and interfaces may also be supported by communications components 614. One or more trained infrared classification systems may be implemented in a remote, real- time environment, as illustrated in FIG. 6B. The infrared classification system 650 may include a thermal imaging camera or other device or system operable to receive and/or generate thermal images and process the received thermal images for input to a trained infrared classification neural network 670. The infrared classification system 650 includes a processor and memory 660, operable to store one or more trained neural networks and implement the neural network run-time interface (such as trained infrared classification neural network 670) thereon. In various embodiments, a training dataset stored in the database 602 may be created from annotated registered image pairs and used to train one or more neural networks and other machine learning algorithms for use in an infrared classification system. Referring to FIG. 6C, an embodiment of a neural network training process will now be described. In one embodiment, the neural network 680 is a convolutional neural network (CNN) that receives the training dataset 682 and outputs a classification for the data. The present disclosure describes a plurality of neural networks that may be trained for one or more infrared image determinations, including but not limited to, detecting and tracking objects and/or people from thermal images, detection of a temperature measurement location in a thermal image, classification of a detected object, and/or determination of whether an individual has a fever. The training dataset includes annotated infrared image data created from registered visual and infrared pairs as described herein. The data may also include annotated infrared data created through other means, including synthetic data generated to simulate real-world images. In one embodiment, the training starts with a forward pass through the neural network 680 including feature extraction, a plurality of convolution layers and pooling layers, a plurality of fully connected layers, and an output layer that includes the desired classification. Next, a backward pass through the neural network 680 may be used to update the CNN parameters in view of errors produced in the forward pass (e.g., misclassified data). In various embodiments, other processes may be used in accordance with the present disclosure. An embodiment for validating the trained neural network is illustrated in FIG.6D. A set of fully annotated validation test data 692 is fed into the trained neural network 690. The validation test data 692 may include annotated infrared data generated from registered infrared/visible image pairs as described herein, that was not used as part of the training dataset 682. Detected errors (e.g., image misclassification) may be analyzed and fed back to the training system to update the training model, which in turn updates the training dataset 682 to create a more accurate classification model. In various embodiments, detected errors may be corrected by adding more examples of the data (e.g., more types of environments), increasing the resolution of the data and/or increasing the accuracy of the thermal modeling, to help distinguish between data types. By adjusting the training dataset to improve accuracy on-the-fly, the operator can avoid costly delays in implementing accurate classification systems. In various embodiments, the system is configured to save data generated in real-time in the field for analysis and training of one or more neural networks. For example, data from a deployed system may be fed back into a CNN training process to refine the CNN to improve classification for a particular environment (e.g., detect temperature of people in an airport), desired classification goal (e.g., train the CNN to detect and track one or more objects) and/or for more accurate performance. Dual-Band Temperature Detection Systems and Methods The various training processes disclosed herein may be used to train infrared imaging systems for a variety of detection and classification tasks. In some embodiments, systems for measuring elevated body temperature (EBT) make use of thermal infrared cameras to measure skin temperature without physical contact--an important aspect of any system used in the presence of people with potentially infectious diseases. A person with a fever will have an elevated core body temperature and under certain conditions, this elevated temperature can manifest itself as elevated skin temperature, particularly on facial skin near the tear duct (canthus), an area of the face with a high degree of blood flow. This canthus surface temperature may only be a few degrees C cooler than the core body temperature. The thermal infrared camera is calibrated to accurately measure the difference between a healthy person and a febrile person may only be a matter of a few degrees C difference in core body temperature. The standard calibration of an infrared camera relates the digital counts for each pixel in the image to the radiance of a surface viewed by those pixels (in the spectral band of the thermal camera). The temperature of the surface can be derived from the radiance measured by the camera by various methods, including a lookup table. In the case of skin temperature measurements, it is possible to calibrate the system directly from digital counts to temperature for two reasons. The first reason is that the range of facial skin temperatures is limited to values from 30 °C to ~38 °C. Over this range of temperature, the in-band radiance for a 7-14 micron longwave camera is nearly linear in temperature, as shown in FIG 7. Referring to FIGs.8A and 8B, an example embodiment of a trained dual-band imaging system will be described, in accordance with one or more embodiments. A dual-band camera 800 includes a first camera 810, a second camera 820, and a beamsplitter 830. The first camera 810 includes image capture and processing components configured to sense and capture a visible image of a field of view. The second camera 820 includes image capture and processing components configured to sense and capture an infrared image of the field of view. The first camera 810 and the second camera 820 are arranged in a camera body 802 and configured to capture the images through a beamsplitter 830 at substantially the same time. The beamsplitter 830 reflects an image of the scene to the second camera 820, while the first camera 810 captures the visible images through the beamsplitter 830. The arrangement disclosed herein allows image pairs to be captured with minimal parallax, which facilitates accurate image processing and classification as described herein. In some systems, a blackbody 840 may be present in the field of view. In some embodiments, the blackbody may be positioned as part of the dual-band camera 800 and/or positioned closed to the camera body 820. The reflection of the blackbody 840 radiation off the beamsplitter 830 allows the black body to positioned physically closer to the thermal camera due to the folded geometry (e.g., black body 840 is (a) + (b) away from the thermal camera 820). In some embodiments, placing blackbody close to the thermal camera allows the black body to appear larger to the thermal camera, and also allows for precision placement of the blackbody at a location to be captured by the thermal camera 820, but outside of the usable image. In various embodiments, the blackbody may include a thermistor, a micro-cavity blackbody (e.g., a long, tapered cone) or other blackbodies. For example, a thermistor, a calibrated device with a resistance value as a function of temperature, can be used as a blackbody by heating the thermistor by passing a current through it, measuring the resistance to get the temperature, which can then be compared to a sensed blackbody temperature from a thermal image. Additional aspect of the present disclosure will now be described with reference to FIG.9. The skin on a person’s face has an in-band emissivity that is very close to the typical emissivity of an area blackbody source, e.g., 0.96. The digital counts measured by an infrared camera looking at a blackbody source and human skin will be the same if the two surface temperatures are the same. An EBT system may have a reference blackbody (or blackbodies) in the field of view at all times, which has the property of embedding a temperature calibration into every thermal infrared image. This also makes it possible to use cameras that do not have built-in calibrations, which may be lower cost to produce because the camera does not require a time-consuming radiometric calibration process. An additional advantage is that atmospheric transmission is folded into the in-scene calibration if the blackbody is located at the same distance from the camera as the test subject. FIG.9 illustrates a thermal image 900 of a test subject 902 with a dual blackbody system in the field of view. A first blackbody 910 has a temperature of 35 C, and a second blackbody 912 has a temperature of 41 C. Temperatures can be interpolated and extrapolated from the digital count values on the two blackbodies, or on a single blackbody if the responsivity of the camera is known. In the case of the single blackbody configuration, it may be necessary to know the responsivity of the camera. Referring to FIG.10, a method 1000 to measure temperature using a single reference blackbody will now be described in accordance with one or more embodiments. In step 1010, the responsivity of the camera is measured using two blackbodies (e.g., at 35C and 40C) with the camera fully warmed up and operating in an environment representative of an expected temperature of operation. The camera software may be configured to provide frame averaging to reduce temporal noise, and the responsivity may be calculated as the difference in digital counts on two regions of interest (ROIs) positioned on the two blackbodies (e.g., 571.8 counts), divided by the temperature difference, which in this example is 5° C. The resulting responsivity is 114 counts/°C. In step 1020, a region of interest is identified on the reference blackbody in the field of view of the camera while viewing a scene of interest which contains test subjects, and the mean digital counts on the reference blackbody are determined. This value is the reference offset. In step 1030, the system measures the digital counts on the target of interest, which may include a region on the face of a test subject (e.g., the hottest part of the face, a specific target location on the face such as a the canthus). The software could automatically detect the person’s face, then locate the hottest spot on the face (e.g., the canthus) and place a spotmeter with a 3x3 pixel size on the centroid of the hot spot to generate the EBT counts measurement. In some embodiments, a visible images is captured and used to identify the object and/or location on the object, and the identified location can then be measured at the corresponding location on the thermal image In step 1040, the reference offset is subtracted from the EBT counts measurement. The system then divides by the responsivity to get a relative temperature difference between the test subject’s canthus and the reference blackbody. This temperature difference is then added to the blackbody temperature. The blackbody temperature could be read out continuously by the software to have an updated measurement. This measurement value could be inserted into the metadata of the thermal image. In step 1050, at the same time, a visible-light image of the test subject is recorded which could be used to further analyze the scene (e.g., help identify the subject). Additional metadata could also be recorded, including the air temperature in the proximity of the measurement site, the relative humidity, the range to the person from the camera (perhaps measured by a stereo camera rig, LIDAR, RADAR or another method). This data could be of additional benefit to determine the atmospheric transmission loss between the camera and the subject, and to correct for the spot size effect, which is a phenomenon whereby small targets appear to be colder than their actual kinetic surface temperature due to stray light in the optics and the modulation transfer function (MTF) of the camera system. Referring to FIG.11, a measurement of camera responsivity will now be described, in accordance with one or more embodiments. A thermal image 1100 is captured by the infrared imaging camera of a field of view. The thermal image 1100 includes an image of a blackbody 1120, which has an associated region of interest (e.g., defined by green box labeled 35C Reference) in the thermal image 1100, from which temperature measurements are taken. A target 1110 is identified in the thermal image 1110 has an associated region of interest (e.g., defined by red box labeled 40C Target) from which temperature measurements for the target are taken. Embodiments of example implementations of a dual-band camera will now be described with reference to FIG.12. The dual-band camera 1200 is configured to reduce system parallax using a dichroic beamsplitter 1220. The dual-band camera 1200 may operate with a trained neural network that was trained using a dual-band camera system and beamsplitter as discussed herein. In the operation, the duel-band camera system 1200 captures video data from an infrared camera 1230 and video data from a visible camera 1210 simultaneously with the optical axes overlaid at a high level of precision. The dual-band camera 1200 uses the beamsplitter 1220 at a 45-degree angle relative to the optical axis of each camera (visible camera 1210 and infrared camera 1230). In various embodiments, the beamsplitter 1230 is coated with indium tin oxide on one side to reflect LWIR radiation with a ~90% efficiency to the infrared camera 1230, while simultaneously transmitting 90% of visible-light to the visible camera 1210. The illustrated system could be implemented in a variety of thermal imaging applications, including EBT applications because there is negligible parallax error between the visible and thermal IR images at all distances. In contrast, a conventional system with two side-by-side image sensors (e.g., boresighted visible and thermal IR cameras side- by-side) will result in a parallax error that becomes more noticeable at closer ranges, because the image pair cannot be registered perfectly in space for all object distances. In some embodiments, a blackbody 1240 is located in a corner of the IR field of view (or other location that does not obstruct or minimally obstructs the captured image). For EBT applications, the dataset could be presented as a visible video of a target person where the system user could mouse over any point on the visible image and see the temperature on each point. This embodiment allows a system operator to intuitively explore different ways to view temperature in an EBT or other temperature measurement implementation. In some embodiments, the visible image may be discarded after object detection, analysis, and temperature measurement is performed to preserve the privacy of the subject (e.g., as may be required by law). The dual-band camera 1200 is controlled by a processing component 1250 (e.g., such as one or more of the processing components of FIGs.5-6B), configured in accordance with a particular use or implementation including synchronization logic 1260 for synchronizing the image capture of the two cameras (e.g., synchronizing shutters and image acquisition for temporal registration of captured images), image processing logic 1262 for processing the captured images into images for use by the system, image pair registration logic 1264 for aligning the image pair (e.g., pixel-by-pixel correspondence), object detection and tracking logic 1266 for identifying one or more target objects in the image pair, and temperature measurement logic 1268 for determining a temperature of an object or portion of an object. In various embodiments, one or more of the logical components may be implemented using image analysis, trained neural networks, statistical analysis, and/or other logic. In some embodiments, the dual-band camera 1200 communicates with a remote processing system 1272 through communications components 1270. The remote processing system 1272 may provide one or more services for the dual-band camera 1200 including training and downloading neural networks, online processing of image pairs, storage of historical data, and other services. An an example of a method of displaying EBT data taken with the beamsplitter system is illustrated in FIG.13. As illustrated, a display 1300 includes the visible image 1310 of an object (e.g., a person), which is overlaid with a thermal image inset 1320 on the visible image 1310 to show temperature data at a selected location (e.g., a location selected by the user by placing a pointer 1330 over a desired region). As illustrated, the hottest part of the test subject’s face is the yellow portion on the canthus. The temperature is 33.1 °C as shown by the color bar on the right side. Referring to FIG.14, the placement of the blackbody close to the IR camera will be in further detail, in accordance with one or more embodiments. The concept of placing a reference blackbody in the scene near the test subject or at the same distance as the test subject has the advantage that atmospheric transmission losses are “basked into” the calibration because both the test subject and the reference blackbody have their infrared signals changed by the same amount: reduced by the transmission of the air path, and increased by the air path signals themselves. However, placing the reference blackbody in the scene at distance has some distinct disadvantages. For one, the camera cannot be panned or tilted over the scene unless the reference blackbody is kept in the field of view and its location is known so that the software can maintain the positioning of an ROI on it. Another reason is that the reference blackbody may be bumped into, obscured by a person or other object or otherwise compromised since it is located in proximity to test subjects, a particularly undesirable situation when there is risk of cross-infection. As illustrated, a system 1400 addresses these issues by placing a reference blackbody 1410 in the field of view of a thermal imaging device 1430 at a close distance to the camera lens. In some embodiments, the blackbody 1410 includes a small circular emitter made of aluminum coated with a high emissivity coating. A heating coil would be used to heat up the emitter and a thermistor or thermocouple could be used to monitor the absolute temperature of the emitter. The blackbody 1410 will be highly out of focus if the blackbody is centimeters from the lens of the camera, while the lens is focused at some convenient working distance (e.g., 3 meters or some similar distance where everything in the scene from 1 meter to infinity would be in sufficient focus for measurement). If the blackbody 1410 is out of focus, then it will not be perceived as being at its actual physical temperature because it is a small target and the radiance from it will be smeared out over more pixels than if it was in focus. A solution to this problem is to place a collimating lens 1420 in front of the blackbody emitter 1410 to focus its rays so that the blackbody appears to be in focus. In another embodiment, a shutter is periodically placed in front of the IR camera lens to expose the sensor to a known temperature scene. The shutter could be instrumented with temperature sensors which would then be used to calculate the reference offset. The shutter could also serve as the flat field correction shutter which would improve the spatial uniformity of the images. Putting the flat field correction shutter in front of the lens also corrects for the non- uniformity introduced by the lens assembly. The disadvantage of this approach is that temperature measurements of test subjects would not have the reference blackbody in the field of view at the same time, reducing traceability. There is also the matter of camera offset drift between activations of the reference shutter. This drift can be mitigated if the cameras are contained in a housing that has thermal mass enough to slow down any variations in ambient air temperature. In various embodiments, systems and methods are provided for correcting for distance- induced temperature errors. The system is configured to correct the apparent temperature of a head based on its distance from the thermal imaging sensors so that someone who is sick does not get counted as healthy just because they are farther away from the camera. There are two effects that are at work to make the temperature measurement decrease with distance. One is atmospheric transmission. The other is the so-called “spot size effect”. The spot size effect manifests itself as a decreasing apparent temperature of a target with decreasing angular size. It is observed that, with respect to the modulation transfer function of the optical system, as a hot target gets smaller, there is a “mixing” of cold pixels at the edges that begins to reduce the apparent temperature of the target. In some embodiments, the spot size effect is pronounced when the target is falls below a threshold size (e.g., is less than 15 by 15 pixels in size). The canthus hot spot on an adult is about 1cm by 1cm. It is observed that for an adult person’s canthus to be at least 15x15 pixels in an image formed by an example camera having a 640 x 512 resolution with a 32-degree horizontal field of view, the target person should be less than 0.8 meters away. For screening of people walking by in a crowd, it may be necessary to measure the distance from the camera to the heads of the targets, perhaps based on the size of the head (assuming adult heads). If we have visible-light images of the same scene, then trained artificial intelligence tools may be used to determine the ages and genders of people and then correct for their head sizes based on estimated age and gender. The distance measurement is used in an air path transmission model that accounts for target distance, air temp (e.g., using FPA temperature or external temperature sensor) and relative humidity (e.g., using a relative humidity sensor). In some embodiments, the air path transmission model is based on Moderate Resolution Atmospheric Transmission (MODTRAN), a radiative transport model developed by the Air Force. Referring to FIGs. 15A and 15B, examples of visible and infrared image pairs will now be described, in accordance with one or more embodiments. Image pair 1500 illustrates an example of a visible-light image 1510 showing fog and a thermal image 1520, showing an image of a man 1522 in the fog. Image pair 1550 illustrates an example of a visible-light image 1560 showing a person behind an obstruction 1570, and a corresponding thermal infrared image 1570. These examples illustrate benefits of a dual image camera system as described herein for both training and detection. Regarding training, the image pairs illustrate different contexts where labeling of visible and infrared images provide different context information for the scene. Object detection, tracking and classification can be improved using both visible-light images and infrared images, such as tracking an object when the visible-light image is obstructed. FIGs.16A, 16B, 16C and 16D are views of an example dual-band camera system, in accordance with one or more embodiments, including a perspective cutaway view (FIG.16A), a side cutaway view (FIG. 16B), a bottom view (FIG.16C), and a perspective view (FIG.16D). An embodiment of a dual-band camera system 1600 includes a housing 1602, an infrared camera 1610, a beamsplitter 1620, a visible-light camera 1630 and optical components 1632 (e.g., a lens assembly) for focusing the visible-light image towards the image capture components of the visible- light camera 1630. FIGs.17A and 17B provide views of another example dual-band camera system, in accordance with one or more embodiments, including a plan view (FIG. 17A) and a front side view (FIG. 17B). The dual-band camera system 1700 includes a housing 1720 with a cover 1712 enclosing the system components, which include an infrared camera 1710 (e.g., a LWIR camera), a beam splitter 1720, a visible-light camera 1730 and processing components 1740 (e.g., a GPU board). The dual-band camera system 1700 is configured to receive infrared waves 1706 and visible- light waves 1708 of a scene. The beamsplitter 1720 allows the visible-light waves 1708 to pass through to the visible-light camera 1730. In various embodiments, the visible-light camera 1730 includes optical components and/or separate optical components 1732 are provided to focus the visible-light 1708 for the image capture components of the visible-light camera 1730. The beamsplitter 1720 is also configured to reflect the infrared waves 1706 through window 1712 towards the infrared camera 1710. Processing components 1740 are provided to warp the infrared images to overlay on the visible-light images and generate a two-channel, dual band video stream. Where applicable, various embodiments provided by the present disclosure can be implemented using hardware, software, or combinations of hardware and software. Also, where applicable, the various hardware components and/or software components set forth herein can be combined into composite components comprising software, hardware, and/or both without departing from the spirit of the present disclosure. Where applicable, the various hardware components and/or software components set forth herein can be separated into sub-components comprising software, hardware, or both without departing from the spirit of the present disclosure. Software in accordance with the present disclosure, such as non-transitory instructions, program code, and/or data, can be stored on one or more non-transitory machine-readable mediums. It is also contemplated that software identified herein can be implemented using one or more general purpose or specific purpose computers and/or computer systems, networked and/or otherwise. Where applicable, the ordering of various steps described herein can be changed, combined into composite steps, and/or separated into sub-steps to provide features described herein. Embodiments described above illustrate but do not limit the invention. It should also be understood that numerous modifications and variations are possible in accordance with the principles of the invention. Accordingly, the scope of the invention is defined only by the following claims.

Claims

CLAIMS What is claimed is: 1. A method comprising: simultaneously capturing a pair of images of a scene comprising an infrared image of the scene and a visible image of the scene; aligning the pair of images so that a pixel location in one of the pair of images has a corresponding pixel location in the other image; analyzing the visible image; analyzing the infrared image using a trained neural network based at least in part on the analyzed visible image.
2. The method of claim 1, wherein simultaneously capturing the pair of images of the scene further comprises capturing the infrared image using an infrared camera and capturing the visible image using a visible-light camera, wherein the infrared camera and the visible-light camera are configured to capture the pair of images of the scene from substantially the same perspective.
3. The method of claim 1, wherein aligning the pair of images so that the pixel location in one of the pair of images has a corresponding pixel location in the other image further comprises determining a transformation between the pair of images so that the pixel location in each image represents the same location in the scene; and applying the transformation to the image pair to align the images.
4. The method of claim 1, wherein analyzing the visible image comprises receiving an object location in the visible image for measurement; and wherein analyzing the infrared image using a trained neural network based at least in part on the analyzed visible image further comprises measuring a temperature of the object location in the infrared image.
5. The method of claim 1, wherein simultaneously capturing the pair of images of the scene further comprises capturing the pair of images of the scene through a beamsplitter arranged to reflect a first image of the scene towards a first camera and pass through a second image of the scene to a second camera.
6. The method of claim 5, wherein the method is performed by a dual-band camera comprising a first blackbody positioned in a field of view of the first camera, and wherein the infrared image includes an image of the first blackbody.
7. The method of claim 6, further comprising measuring a responsivity of the first camera using the first blackbody and a second blackbody disposed in the field of view.
8. The method of claim 6, wherein analyzing the infrared image using a trained neural network based at least in part on the analyzed visible image further comprises calculating a temperature based on a reference offset and a temperature measurement of the first blackbody.
9 The method of claim 6, wherein the first blackbody is a thermistor.
10. The method of claim 6, wherein the first black body comprises a collimating lens.
11. A system comprising: an infrared camera configured to capture an infrared image of a scene; a visible-light camera configured to capture a visible-light image of the scene; and a logic device configured to: simultaneously capture a pair of images of the scene comprising the infrared image of the scene and the visible image of the scene; align the pair of images so that a pixel location in one of the pair of images has a corresponding pixel location in the other image; classify the visible image; analyze the infrared image based, at least in part, on the classification of the visible image.
12. The system of claim 11, wherein the infrared camera and the visible-light camera are configured to capture the pair of images of the scene from substantially the same perspective.
13. The system of claim 11, wherein the logic device is further configured to determine a transformation between the pair of images so that the pixel location in each image represents the same location in the scene; and apply the transformation to the image pair to align the images.
14. The system of claim 11, wherein the logic device is further configured to receive an object location in the visible image for measurement; and analyze the infrared image using a trained neural network based at least in part on the analyzed visible image to measure a temperature of the object location in the infrared image.
15. The system of claim 11, further comprising a beamsplitter arranged to reflect a first image of the scene towards the infrared camera and pass through a second image of the scene to the visible-light camera.
16. The system of claim 15, further comprising a first blackbody attached to the system and positioned in a field of view of the infrared camera, and wherein the infrared image includes an image of the first blackbody.
17. The system of claim 16, wherein the logic device is further configured to measure a responsivity of the infrared camera using the first blackbody and a second blackbody disposed in the field of view.
18. The system of claim 16, wherein the logic device is further configured to calculate a temperature based on a reference offset and a temperature measurement of the first blackbody.
19. The system of claim 16, wherein the first blackbody comprises a thermistor.
20. The system of claim 16, wherein the first black body comprises a collimating lens.
PCT/US2021/031435 2020-05-08 2021-05-07 Dual-band temperature detection systems and methods WO2021226554A1 (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/981,337 US20230054197A1 (en) 2020-05-08 2022-11-04 Dual-band temperature detection systems and methods

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US202063022281P 2020-05-08 2020-05-08
US63/022,281 2020-05-08

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/981,337 Continuation US20230054197A1 (en) 2020-05-08 2022-11-04 Dual-band temperature detection systems and methods

Publications (1)

Publication Number Publication Date
WO2021226554A1 true WO2021226554A1 (en) 2021-11-11

Family

ID=76181294

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2021/031435 WO2021226554A1 (en) 2020-05-08 2021-05-07 Dual-band temperature detection systems and methods

Country Status (2)

Country Link
US (1) US20230054197A1 (en)
WO (1) WO2021226554A1 (en)

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
TWI756932B (en) * 2020-11-20 2022-03-01 財團法人工業技術研究院 Thermal imaging apparatus and temperature correction method of thermal imager
US20230298352A1 (en) * 2022-03-16 2023-09-21 Meta Platforms Technologies, Llc Remote sensing security and communication system

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060261271A1 (en) * 2005-05-17 2006-11-23 Tzong-Sheng Lee Dual-band reflective infrared thermal imaging system
US20160316154A1 (en) * 2014-01-05 2016-10-27 Flir Systems, Inc. Device attachment with dual band imaging sensor
CN110044486A (en) * 2019-03-11 2019-07-23 武汉高德智感科技有限公司 Method, apparatus, the equipment of repetition of alarms are avoided for human body inspection and quarantine system
CN110110629A (en) * 2019-04-25 2019-08-09 北京工业大学 Personal information detection method and system towards indoor environmental condition control

Family Cites Families (17)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7483213B2 (en) * 2006-03-24 2009-01-27 Omnitech Partners Image combining viewer
US9681066B2 (en) * 2013-07-08 2017-06-13 Flir Systems Ab Facilitating improved calibration of captured infrared data values by an IR imaging system in a thermography arrangement
US9386222B2 (en) * 2014-06-20 2016-07-05 Qualcomm Incorporated Multi-camera system using folded optics free from parallax artifacts
US10648960B2 (en) * 2015-05-29 2020-05-12 Rebellion Photonics, Inc. Hydrogen sulfide imaging system
US20180173933A1 (en) * 2016-12-16 2018-06-21 Qualcomm Incorporated User authentication using iris sector
US10520742B1 (en) * 2017-02-13 2019-12-31 Facebook Technologies, Llc Beamsplitter assembly for eye tracking in head-mounted displays
US10580130B2 (en) * 2017-03-24 2020-03-03 Curadel, LLC Tissue identification by an imaging system using color information
US10524655B2 (en) * 2017-09-27 2020-01-07 International Business Machines Corporation Ophthalmoscope using natural pupil dilation
US10579897B2 (en) * 2017-10-02 2020-03-03 Xnor.ai Inc. Image based object detection
US10679443B2 (en) * 2017-10-13 2020-06-09 Alcatraz AI, Inc. System and method for controlling access to a building with facial recognition
US11100350B2 (en) * 2018-02-19 2021-08-24 Avigilon Corporation Method and system for object classification using visible and invisible light images
CN108876833A (en) * 2018-03-29 2018-11-23 北京旷视科技有限公司 Image processing method, image processing apparatus and computer readable storage medium
US20200050902A1 (en) * 2018-08-09 2020-02-13 Electronics And Telecommunications Research Institute Method for analyzing material using neural network based on multimodal input and apparatus using the same
US10880472B1 (en) * 2019-06-12 2020-12-29 Air Products And Chemicals, Inc. Smart sensor measurement system
US11037312B2 (en) * 2019-06-29 2021-06-15 Intel Corporation Technologies for thermal enhanced semantic segmentation of two-dimensional images
KR20220163998A (en) * 2020-04-06 2022-12-12 텔레다인 플리어, 엘엘시 Non-Contact Temperature Measurement in Thermal Imaging Systems and Methods
US11875544B2 (en) * 2020-04-30 2024-01-16 Teledyne Flir Commercial Systems, Inc. Annotation of infrared images for machine learning using beamsplitter-based camera system and methods

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060261271A1 (en) * 2005-05-17 2006-11-23 Tzong-Sheng Lee Dual-band reflective infrared thermal imaging system
US20160316154A1 (en) * 2014-01-05 2016-10-27 Flir Systems, Inc. Device attachment with dual band imaging sensor
CN110044486A (en) * 2019-03-11 2019-07-23 武汉高德智感科技有限公司 Method, apparatus, the equipment of repetition of alarms are avoided for human body inspection and quarantine system
CN110110629A (en) * 2019-04-25 2019-08-09 北京工业大学 Personal information detection method and system towards indoor environmental condition control

Also Published As

Publication number Publication date
US20230054197A1 (en) 2023-02-23

Similar Documents

Publication Publication Date Title
US10129490B2 (en) Systems and approaches for thermal image corrections
US20230079693A1 (en) Elevated temperature screening systems and methods
US7613360B2 (en) Multi-spectral fusion for video surveillance
US20230043342A1 (en) Non-contact temperature measurement in thermal imaging systems and methods
EP3265996B1 (en) Quantifying gas in passive optical gas imaging
US20230054197A1 (en) Dual-band temperature detection systems and methods
US9762871B2 (en) Camera assisted two dimensional keystone correction
US9684963B2 (en) Systems and methods for dynamic registration of multimodal images
US9681066B2 (en) Facilitating improved calibration of captured infrared data values by an IR imaging system in a thermography arrangement
US20110122251A1 (en) Comparison of Infrared Images
CN104272717A (en) A method and system for performing alignment of a projection image to detected infrared (IR) radiation information
US10735659B2 (en) Rotation-adaptive video analytics camera and method
US20120086810A1 (en) Method for a touchless determination of the temperature of an object and corresponding thermal imaging camera
WO2021259365A1 (en) Target temperature measurement method and apparatus, and temperature measurement system
CN110675448A (en) Ground light remote sensing monitoring method, system and storage medium based on civil aircraft
US12105020B2 (en) Gas lens filter systems and methods
US11875544B2 (en) Annotation of infrared images for machine learning using beamsplitter-based camera system and methods
CN114485953A (en) Temperature measuring method, device and system
CN112771568A (en) Infrared image processing method, device, movable platform and computer readable medium
KR102251307B1 (en) Thermal camera system with distance measuring function
US20240256977A1 (en) Image setting determination and associated machine learning in infrared imaging systems and methods
CN113272855B (en) Response normalization for overlapping multiple image applications
KR102487590B1 (en) Method for measuring of object based on face-recognition
Efimov et al. Digital multispectral images superimposition based on preliminary calibration or contour analysis: advantages and disadvantages
KR102593382B1 (en) Method for measuring temperature of infrared radiator for thermal camera calibration

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21728770

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21728770

Country of ref document: EP

Kind code of ref document: A1